[
  {
    "id": "de-conversation-01-base17-block-free-n1",
    "case": "de-conversation-01",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 1,
    "block_size": 4,
    "plain": "Ich habe Humaneness Podcast gehört, eine Feinabstimmung von Humaneness Voice Large. Du klingst verdächtig begeistert. Hat dich ein Sprachmodell wirklich überrascht? Es liest ein Skript mit mehreren Stimmen und eigenen Spielanweisungen pro Satz. Moment, eine Stimme darf dir misstrauen, während die andere schon jubelt? Genau. Wir testen vierzig Emotionsklassen und behaupten nicht, dass jede perfekt gelingt. Diese Einschränkung macht die Vorführung irgendwie glaubwürdiger. Ärgerlich vernünftig von dir. Im Training kommen echte Gespräche und Schauspielbeispiele zusammen, damit beide Formate erhalten bleiben. Na gut. Zeig mir diese kleine Pause, bevor jemand widerwillig zugibt, beeindruckt zu sein. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/e8866980c7745ba72d40527d6dcaf5bc59d2ad1841a0fc97317a515fe7df3615.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-01-base17-block-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 840960
      },
      {
        "candidate_id": "de-conversation-01-base17-block-free-u01-c00",
        "output_offset_sample": 856320,
        "first_word_output_sample": 860160,
        "last_word_output_sample": 1758720
      },
      {
        "candidate_id": "de-conversation-01-base17-block-free-u02-c00",
        "output_offset_sample": 1774080,
        "first_word_output_sample": 1777920,
        "last_word_output_sample": 2707200
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 56.55,
    "samples": 2714400,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-01-base17-block-free-n1.mp3"
  },
  {
    "id": "de-conversation-01-base17-block-free-n4",
    "case": "de-conversation-01",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 4,
    "block_size": 4,
    "plain": "Ich habe Humaneness Podcast gehört, eine Feinabstimmung von Humaneness Voice Large. Du klingst verdächtig begeistert. Hat dich ein Sprachmodell wirklich überrascht? Es liest ein Skript mit mehreren Stimmen und eigenen Spielanweisungen pro Satz. Moment, eine Stimme darf dir misstrauen, während die andere schon jubelt? Genau. Wir testen vierzig Emotionsklassen und behaupten nicht, dass jede perfekt gelingt. Diese Einschränkung macht die Vorführung irgendwie glaubwürdiger. Ärgerlich vernünftig von dir. Im Training kommen echte Gespräche und Schauspielbeispiele zusammen, damit beide Formate erhalten bleiben. Na gut. Zeig mir diese kleine Pause, bevor jemand widerwillig zugibt, beeindruckt zu sein. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/7e0dd0e88613c0fc6578e46f7f121b671ad29bf2ffed068aa361a66a3993ba0c.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-01-base17-block-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 840960
      },
      {
        "candidate_id": "de-conversation-01-base17-block-free-u01-c01",
        "output_offset_sample": 856320,
        "first_word_output_sample": 860160,
        "last_word_output_sample": 1820160
      },
      {
        "candidate_id": "de-conversation-01-base17-block-free-u02-c01",
        "output_offset_sample": 1839360,
        "first_word_output_sample": 1839360,
        "last_word_output_sample": 2772480
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 57.91,
    "samples": 2779680,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-01-base17-block-free-n4.mp3"
  },
  {
    "id": "de-conversation-01-base17-block-free-n8",
    "case": "de-conversation-01",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 8,
    "block_size": 4,
    "plain": "Ich habe Humaneness Podcast gehört, eine Feinabstimmung von Humaneness Voice Large. Du klingst verdächtig begeistert. Hat dich ein Sprachmodell wirklich überrascht? Es liest ein Skript mit mehreren Stimmen und eigenen Spielanweisungen pro Satz. Moment, eine Stimme darf dir misstrauen, während die andere schon jubelt? Genau. Wir testen vierzig Emotionsklassen und behaupten nicht, dass jede perfekt gelingt. Diese Einschränkung macht die Vorführung irgendwie glaubwürdiger. Ärgerlich vernünftig von dir. Im Training kommen echte Gespräche und Schauspielbeispiele zusammen, damit beide Formate erhalten bleiben. Na gut. Zeig mir diese kleine Pause, bevor jemand widerwillig zugibt, beeindruckt zu sein. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/7e0dd0e88613c0fc6578e46f7f121b671ad29bf2ffed068aa361a66a3993ba0c.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-01-base17-block-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 840960
      },
      {
        "candidate_id": "de-conversation-01-base17-block-free-u01-c01",
        "output_offset_sample": 856320,
        "first_word_output_sample": 860160,
        "last_word_output_sample": 1820160
      },
      {
        "candidate_id": "de-conversation-01-base17-block-free-u02-c01",
        "output_offset_sample": 1839360,
        "first_word_output_sample": 1839360,
        "last_word_output_sample": 2772480
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 57.91,
    "samples": 2779680,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-01-base17-block-free-n8.mp3"
  },
  {
    "id": "de-conversation-01-base17-block-timed-n1",
    "case": "de-conversation-01",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 1,
    "block_size": 4,
    "plain": "Ich habe Humaneness Podcast gehört, eine Feinabstimmung von Humaneness Voice Large. Du klingst verdächtig begeistert. Hat dich ein Sprachmodell wirklich überrascht? Es liest ein Skript mit mehreren Stimmen und eigenen Spielanweisungen pro Satz. Moment, eine Stimme darf dir misstrauen, während die andere schon jubelt? Genau. Wir testen vierzig Emotionsklassen und behaupten nicht, dass jede perfekt gelingt. Diese Einschränkung macht die Vorführung irgendwie glaubwürdiger. Ärgerlich vernünftig von dir. Im Training kommen echte Gespräche und Schauspielbeispiele zusammen, damit beide Formate erhalten bleiben. Na gut. Zeig mir diese kleine Pause, bevor jemand widerwillig zugibt, beeindruckt zu sein. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/3a54c1eb4c2d8e78d4a4c2f3604d014745d42005a19ea9abf8069f6251bc0616.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-01-base17-block-timed-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 748800
      },
      {
        "candidate_id": "de-conversation-01-base17-block-timed-u01-c00",
        "output_offset_sample": 764160,
        "first_word_output_sample": 768000,
        "last_word_output_sample": 1620480
      },
      {
        "candidate_id": "de-conversation-01-base17-block-timed-u02-c00",
        "output_offset_sample": 1635840,
        "first_word_output_sample": 1639680,
        "last_word_output_sample": 2588160
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 54.07,
    "samples": 2595360,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-01-base17-block-timed-n1.mp3"
  },
  {
    "id": "de-conversation-01-base17-block-timed-n4",
    "case": "de-conversation-01",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 4,
    "block_size": 4,
    "plain": "Ich habe Humaneness Podcast gehört, eine Feinabstimmung von Humaneness Voice Large. Du klingst verdächtig begeistert. Hat dich ein Sprachmodell wirklich überrascht? Es liest ein Skript mit mehreren Stimmen und eigenen Spielanweisungen pro Satz. Moment, eine Stimme darf dir misstrauen, während die andere schon jubelt? Genau. Wir testen vierzig Emotionsklassen und behaupten nicht, dass jede perfekt gelingt. Diese Einschränkung macht die Vorführung irgendwie glaubwürdiger. Ärgerlich vernünftig von dir. Im Training kommen echte Gespräche und Schauspielbeispiele zusammen, damit beide Formate erhalten bleiben. Na gut. Zeig mir diese kleine Pause, bevor jemand widerwillig zugibt, beeindruckt zu sein. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/1ba0ab99b68321859273f1e2a06e995d2dd48a189d57aff6eb2dd3346124e677.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-01-base17-block-timed-u00-c01",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 748800
      },
      {
        "candidate_id": "de-conversation-01-base17-block-timed-u01-c02",
        "output_offset_sample": 764160,
        "first_word_output_sample": 768000,
        "last_word_output_sample": 1608960
      },
      {
        "candidate_id": "de-conversation-01-base17-block-timed-u02-c01",
        "output_offset_sample": 1624320,
        "first_word_output_sample": 1628160,
        "last_word_output_sample": 2572800
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 53.75,
    "samples": 2580000,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-01-base17-block-timed-n4.mp3"
  },
  {
    "id": "de-conversation-01-base17-block-timed-n8",
    "case": "de-conversation-01",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 8,
    "block_size": 4,
    "plain": "Ich habe Humaneness Podcast gehört, eine Feinabstimmung von Humaneness Voice Large. Du klingst verdächtig begeistert. Hat dich ein Sprachmodell wirklich überrascht? Es liest ein Skript mit mehreren Stimmen und eigenen Spielanweisungen pro Satz. Moment, eine Stimme darf dir misstrauen, während die andere schon jubelt? Genau. Wir testen vierzig Emotionsklassen und behaupten nicht, dass jede perfekt gelingt. Diese Einschränkung macht die Vorführung irgendwie glaubwürdiger. Ärgerlich vernünftig von dir. Im Training kommen echte Gespräche und Schauspielbeispiele zusammen, damit beide Formate erhalten bleiben. Na gut. Zeig mir diese kleine Pause, bevor jemand widerwillig zugibt, beeindruckt zu sein. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/3cdd3bca92b0ca0f85229b3b4694dee879cc5a9a242d2d70f669e7f1d4e312c1.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-01-base17-block-timed-u00-c07",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 748800
      },
      {
        "candidate_id": "de-conversation-01-base17-block-timed-u01-c02",
        "output_offset_sample": 764160,
        "first_word_output_sample": 768000,
        "last_word_output_sample": 1608960
      },
      {
        "candidate_id": "de-conversation-01-base17-block-timed-u02-c01",
        "output_offset_sample": 1624320,
        "first_word_output_sample": 1628160,
        "last_word_output_sample": 2572800
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 53.75,
    "samples": 2580000,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-01-base17-block-timed-n8.mp3"
  },
  {
    "id": "de-conversation-01-base17-segment-free-n1",
    "case": "de-conversation-01",
    "model": "base17",
    "strategy": "segment",
    "timing": "free",
    "best_of": 1,
    "block_size": 1,
    "plain": "Ich habe Humaneness Podcast gehört, eine Feinabstimmung von Humaneness Voice Large. Du klingst verdächtig begeistert. Hat dich ein Sprachmodell wirklich überrascht? Es liest ein Skript mit mehreren Stimmen und eigenen Spielanweisungen pro Satz. Moment, eine Stimme darf dir misstrauen, während die andere schon jubelt? Genau. Wir testen vierzig Emotionsklassen und behaupten nicht, dass jede perfekt gelingt. Diese Einschränkung macht die Vorführung irgendwie glaubwürdiger. Ärgerlich vernünftig von dir. Im Training kommen echte Gespräche und Schauspielbeispiele zusammen, damit beide Formate erhalten bleiben. Na gut. Zeig mir diese kleine Pause, bevor jemand widerwillig zugibt, beeindruckt zu sein. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": null,
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-01-base17-segment-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 380160
      },
      {
        "candidate_id": "de-conversation-01-base17-segment-free-u01-c00",
        "output_offset_sample": 399360,
        "first_word_output_sample": 399360,
        "last_word_output_sample": 698880
      },
      {
        "candidate_id": "de-conversation-01-base17-segment-free-u02-c00",
        "output_offset_sample": 718080,
        "first_word_output_sample": 718080,
        "last_word_output_sample": 990720
      },
      {
        "candidate_id": "de-conversation-01-base17-segment-free-u03-c00",
        "output_offset_sample": 1006080,
        "first_word_output_sample": 1009920,
        "last_word_output_sample": 1251840
      },
      {
        "candidate_id": "de-conversation-01-base17-segment-free-u04-c00",
        "output_offset_sample": 1267200,
        "first_word_output_sample": 1271040,
        "last_word_output_sample": 1608960
      },
      {
        "candidate_id": "de-conversation-01-base17-segment-free-u05-c00",
        "output_offset_sample": 1628160,
        "first_word_output_sample": 1628160,
        "last_word_output_sample": 1962240
      },
      {
        "candidate_id": "de-conversation-01-base17-segment-free-u06-c00",
        "output_offset_sample": 1981440,
        "first_word_output_sample": 1981440,
        "last_word_output_sample": 2296320
      },
      {
        "candidate_id": "de-conversation-01-base17-segment-free-u07-c00",
        "output_offset_sample": 2311680,
        "first_word_output_sample": 2315520,
        "last_word_output_sample": 2641920
      },
      {
        "candidate_id": "de-conversation-01-base17-segment-free-u08-c00",
        "output_offset_sample": 2657280,
        "first_word_output_sample": 2661120,
        "last_word_output_sample": 3160320
      },
      {
        "candidate_id": "de-conversation-01-base17-segment-free-u09-c00",
        "output_offset_sample": 3175680,
        "first_word_output_sample": 3179520,
        "last_word_output_sample": 3467520
      },
      {
        "candidate_id": "de-conversation-01-base17-segment-free-u10-c00",
        "output_offset_sample": 3486720,
        "first_word_output_sample": 3486720,
        "last_word_output_sample": 3797760
      },
      {
        "candidate_id": "de-conversation-01-base17-segment-free-u11-c00",
        "output_offset_sample": 3813120,
        "first_word_output_sample": 3816960,
        "last_word_output_sample": 4143360
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 86.47,
    "samples": 4150560,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-01-base17-segment-free-n1.mp3"
  },
  {
    "id": "de-conversation-01-base17-segment-timed-n1",
    "case": "de-conversation-01",
    "model": "base17",
    "strategy": "segment",
    "timing": "timed",
    "best_of": 1,
    "block_size": 1,
    "plain": "Ich habe Humaneness Podcast gehört, eine Feinabstimmung von Humaneness Voice Large. Du klingst verdächtig begeistert. Hat dich ein Sprachmodell wirklich überrascht? Es liest ein Skript mit mehreren Stimmen und eigenen Spielanweisungen pro Satz. Moment, eine Stimme darf dir misstrauen, während die andere schon jubelt? Genau. Wir testen vierzig Emotionsklassen und behaupten nicht, dass jede perfekt gelingt. Diese Einschränkung macht die Vorführung irgendwie glaubwürdiger. Ärgerlich vernünftig von dir. Im Training kommen echte Gespräche und Schauspielbeispiele zusammen, damit beide Formate erhalten bleiben. Na gut. Zeig mir diese kleine Pause, bevor jemand widerwillig zugibt, beeindruckt zu sein. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/49ce51dcd1d79fe72b1bb7abc7283c8ba36cb0bcf9646172294ad52fed888a5c.mp3",
    "review_candidates": [
      "de-conversation-01-base17-segment-timed-u00-c00"
    ]
  },
  {
    "id": "de-conversation-01-podcast-block-free-n1",
    "case": "de-conversation-01",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 1,
    "block_size": 4,
    "plain": "Ich habe Humaneness Podcast gehört, eine Feinabstimmung von Humaneness Voice Large. Du klingst verdächtig begeistert. Hat dich ein Sprachmodell wirklich überrascht? Es liest ein Skript mit mehreren Stimmen und eigenen Spielanweisungen pro Satz. Moment, eine Stimme darf dir misstrauen, während die andere schon jubelt? Genau. Wir testen vierzig Emotionsklassen und behaupten nicht, dass jede perfekt gelingt. Diese Einschränkung macht die Vorführung irgendwie glaubwürdiger. Ärgerlich vernünftig von dir. Im Training kommen echte Gespräche und Schauspielbeispiele zusammen, damit beide Formate erhalten bleiben. Na gut. Zeig mir diese kleine Pause, bevor jemand widerwillig zugibt, beeindruckt zu sein. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/9cc80633f3fdcded7308fadfafaded99bf4a647eccf8bc4963e8fb070c6ed219.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-01-podcast-block-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 867840
      },
      {
        "candidate_id": "de-conversation-01-podcast-block-free-u01-c00",
        "output_offset_sample": 883200,
        "first_word_output_sample": 887040,
        "last_word_output_sample": 1941120
      },
      {
        "candidate_id": "de-conversation-01-podcast-block-free-u02-c00",
        "output_offset_sample": 1960320,
        "first_word_output_sample": 1960320,
        "last_word_output_sample": 2977920
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 62.19,
    "samples": 2985120,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-01-podcast-block-free-n1.mp3"
  },
  {
    "id": "de-conversation-01-podcast-block-free-n4",
    "case": "de-conversation-01",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 4,
    "block_size": 4,
    "plain": "Ich habe Humaneness Podcast gehört, eine Feinabstimmung von Humaneness Voice Large. Du klingst verdächtig begeistert. Hat dich ein Sprachmodell wirklich überrascht? Es liest ein Skript mit mehreren Stimmen und eigenen Spielanweisungen pro Satz. Moment, eine Stimme darf dir misstrauen, während die andere schon jubelt? Genau. Wir testen vierzig Emotionsklassen und behaupten nicht, dass jede perfekt gelingt. Diese Einschränkung macht die Vorführung irgendwie glaubwürdiger. Ärgerlich vernünftig von dir. Im Training kommen echte Gespräche und Schauspielbeispiele zusammen, damit beide Formate erhalten bleiben. Na gut. Zeig mir diese kleine Pause, bevor jemand widerwillig zugibt, beeindruckt zu sein. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/3197f78f36a2bf8800a9fbc414e30fb6678efdb1fc1c4840cd87b323b33d26e3.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-01-podcast-block-free-u00-c02",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 848640
      },
      {
        "candidate_id": "de-conversation-01-podcast-block-free-u01-c03",
        "output_offset_sample": 867840,
        "first_word_output_sample": 867840,
        "last_word_output_sample": 1854720
      },
      {
        "candidate_id": "de-conversation-01-podcast-block-free-u02-c02",
        "output_offset_sample": 1873920,
        "first_word_output_sample": 1873920,
        "last_word_output_sample": 2853120
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 59.59,
    "samples": 2860320,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-01-podcast-block-free-n4.mp3"
  },
  {
    "id": "de-conversation-01-podcast-block-free-n8",
    "case": "de-conversation-01",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 8,
    "block_size": 4,
    "plain": "Ich habe Humaneness Podcast gehört, eine Feinabstimmung von Humaneness Voice Large. Du klingst verdächtig begeistert. Hat dich ein Sprachmodell wirklich überrascht? Es liest ein Skript mit mehreren Stimmen und eigenen Spielanweisungen pro Satz. Moment, eine Stimme darf dir misstrauen, während die andere schon jubelt? Genau. Wir testen vierzig Emotionsklassen und behaupten nicht, dass jede perfekt gelingt. Diese Einschränkung macht die Vorführung irgendwie glaubwürdiger. Ärgerlich vernünftig von dir. Im Training kommen echte Gespräche und Schauspielbeispiele zusammen, damit beide Formate erhalten bleiben. Na gut. Zeig mir diese kleine Pause, bevor jemand widerwillig zugibt, beeindruckt zu sein. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/85dc68eaeeb37cf35fb9809546f124ddf3c326c8c44ffad391a9e1578cce9361.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-01-podcast-block-free-u00-c05",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 833280
      },
      {
        "candidate_id": "de-conversation-01-podcast-block-free-u01-c03",
        "output_offset_sample": 852480,
        "first_word_output_sample": 852480,
        "last_word_output_sample": 1839360
      },
      {
        "candidate_id": "de-conversation-01-podcast-block-free-u02-c07",
        "output_offset_sample": 1858560,
        "first_word_output_sample": 1858560,
        "last_word_output_sample": 2833920
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 59.19,
    "samples": 2841120,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-01-podcast-block-free-n8.mp3"
  },
  {
    "id": "de-conversation-01-podcast-block-timed-n1",
    "case": "de-conversation-01",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 1,
    "block_size": 4,
    "plain": "Ich habe Humaneness Podcast gehört, eine Feinabstimmung von Humaneness Voice Large. Du klingst verdächtig begeistert. Hat dich ein Sprachmodell wirklich überrascht? Es liest ein Skript mit mehreren Stimmen und eigenen Spielanweisungen pro Satz. Moment, eine Stimme darf dir misstrauen, während die andere schon jubelt? Genau. Wir testen vierzig Emotionsklassen und behaupten nicht, dass jede perfekt gelingt. Diese Einschränkung macht die Vorführung irgendwie glaubwürdiger. Ärgerlich vernünftig von dir. Im Training kommen echte Gespräche und Schauspielbeispiele zusammen, damit beide Formate erhalten bleiben. Na gut. Zeig mir diese kleine Pause, bevor jemand widerwillig zugibt, beeindruckt zu sein. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/79250e8e3f0633519e9f29617818174a079b19ea043fcdef1c299545491d758c.mp3",
    "review_candidates": [
      "de-conversation-01-podcast-block-timed-u01-c00",
      "de-conversation-01-podcast-block-timed-u02-c00"
    ]
  },
  {
    "id": "de-conversation-01-podcast-block-timed-n4",
    "case": "de-conversation-01",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 4,
    "block_size": 4,
    "plain": "Ich habe Humaneness Podcast gehört, eine Feinabstimmung von Humaneness Voice Large. Du klingst verdächtig begeistert. Hat dich ein Sprachmodell wirklich überrascht? Es liest ein Skript mit mehreren Stimmen und eigenen Spielanweisungen pro Satz. Moment, eine Stimme darf dir misstrauen, während die andere schon jubelt? Genau. Wir testen vierzig Emotionsklassen und behaupten nicht, dass jede perfekt gelingt. Diese Einschränkung macht die Vorführung irgendwie glaubwürdiger. Ärgerlich vernünftig von dir. Im Training kommen echte Gespräche und Schauspielbeispiele zusammen, damit beide Formate erhalten bleiben. Na gut. Zeig mir diese kleine Pause, bevor jemand widerwillig zugibt, beeindruckt zu sein. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/37ac1368ad1d96a84f8a2bd23f5cfcaf03b9f4fb54479a1521e3213e9a2b2da8.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-01-podcast-block-timed-u00-c01",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 748800
      },
      {
        "candidate_id": "de-conversation-01-podcast-block-timed-u01-c03",
        "output_offset_sample": 768000,
        "first_word_output_sample": 768000,
        "last_word_output_sample": 1616640
      },
      {
        "candidate_id": "de-conversation-01-podcast-block-timed-u02-c03",
        "output_offset_sample": 1635840,
        "first_word_output_sample": 1635840,
        "last_word_output_sample": 2592000
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 54.15,
    "samples": 2599200,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-01-podcast-block-timed-n4.mp3"
  },
  {
    "id": "de-conversation-01-podcast-block-timed-n8",
    "case": "de-conversation-01",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 8,
    "block_size": 4,
    "plain": "Ich habe Humaneness Podcast gehört, eine Feinabstimmung von Humaneness Voice Large. Du klingst verdächtig begeistert. Hat dich ein Sprachmodell wirklich überrascht? Es liest ein Skript mit mehreren Stimmen und eigenen Spielanweisungen pro Satz. Moment, eine Stimme darf dir misstrauen, während die andere schon jubelt? Genau. Wir testen vierzig Emotionsklassen und behaupten nicht, dass jede perfekt gelingt. Diese Einschränkung macht die Vorführung irgendwie glaubwürdiger. Ärgerlich vernünftig von dir. Im Training kommen echte Gespräche und Schauspielbeispiele zusammen, damit beide Formate erhalten bleiben. Na gut. Zeig mir diese kleine Pause, bevor jemand widerwillig zugibt, beeindruckt zu sein. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/6565295e092ad49acb7cf9d6205d08068bd89a665d7b864d1a0e3a823fc88104.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-01-podcast-block-timed-u00-c04",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 748800
      },
      {
        "candidate_id": "de-conversation-01-podcast-block-timed-u01-c03",
        "output_offset_sample": 768000,
        "first_word_output_sample": 768000,
        "last_word_output_sample": 1616640
      },
      {
        "candidate_id": "de-conversation-01-podcast-block-timed-u02-c05",
        "output_offset_sample": 1635840,
        "first_word_output_sample": 1635840,
        "last_word_output_sample": 2588160
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 54.07,
    "samples": 2595360,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-01-podcast-block-timed-n8.mp3"
  },
  {
    "id": "de-conversation-01-podcast-segment-free-n1",
    "case": "de-conversation-01",
    "model": "podcast",
    "strategy": "segment",
    "timing": "free",
    "best_of": 1,
    "block_size": 1,
    "plain": "Ich habe Humaneness Podcast gehört, eine Feinabstimmung von Humaneness Voice Large. Du klingst verdächtig begeistert. Hat dich ein Sprachmodell wirklich überrascht? Es liest ein Skript mit mehreren Stimmen und eigenen Spielanweisungen pro Satz. Moment, eine Stimme darf dir misstrauen, während die andere schon jubelt? Genau. Wir testen vierzig Emotionsklassen und behaupten nicht, dass jede perfekt gelingt. Diese Einschränkung macht die Vorführung irgendwie glaubwürdiger. Ärgerlich vernünftig von dir. Im Training kommen echte Gespräche und Schauspielbeispiele zusammen, damit beide Formate erhalten bleiben. Na gut. Zeig mir diese kleine Pause, bevor jemand widerwillig zugibt, beeindruckt zu sein. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/a571521833b2f72eb94c048cb771ac7f51ac8478dece179d2f3fedf371ace2b6.mp3",
    "review_candidates": [
      "de-conversation-01-podcast-segment-free-u10-c00"
    ]
  },
  {
    "id": "de-conversation-01-podcast-segment-timed-n1",
    "case": "de-conversation-01",
    "model": "podcast",
    "strategy": "segment",
    "timing": "timed",
    "best_of": 1,
    "block_size": 1,
    "plain": "Ich habe Humaneness Podcast gehört, eine Feinabstimmung von Humaneness Voice Large. Du klingst verdächtig begeistert. Hat dich ein Sprachmodell wirklich überrascht? Es liest ein Skript mit mehreren Stimmen und eigenen Spielanweisungen pro Satz. Moment, eine Stimme darf dir misstrauen, während die andere schon jubelt? Genau. Wir testen vierzig Emotionsklassen und behaupten nicht, dass jede perfekt gelingt. Diese Einschränkung macht die Vorführung irgendwie glaubwürdiger. Ärgerlich vernünftig von dir. Im Training kommen echte Gespräche und Schauspielbeispiele zusammen, damit beide Formate erhalten bleiben. Na gut. Zeig mir diese kleine Pause, bevor jemand widerwillig zugibt, beeindruckt zu sein. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/ab40704c842d57d8558e0dcd2bbd9e6f73d5953b97000ad9e7cd29ecfad27c34.mp3",
    "review_candidates": [
      "de-conversation-01-podcast-segment-timed-u00-c00",
      "de-conversation-01-podcast-segment-timed-u02-c00"
    ]
  },
  {
    "id": "de-conversation-02-base17-block-free-n1",
    "case": "de-conversation-02",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 1,
    "block_size": 5,
    "plain": "Stell dir ein winziges Radiostudio vor, in dem zwei Leute dieses Modell entdecken. Bitte sag mir, dass dort eine völlig übertriebene rote Aufnahmelampe leuchtet. Natürlich. Humaneness Voice Large liefert die Grundlage, das Podcasttraining übt den Dialog. Darf die zweite Stimme auch skeptisch nachfragen, statt immer nur freundlich zuzustimmen? Unbedingt. Jede Äußerung bekommt ihren eigenen Text und eine eigene emotionale Richtung. Von einer begeisterten Beschreibung lasse ich mich noch lange nicht überzeugen. Ich will es hören. Deshalb stehen die Referenzclips direkt neben den Ausgaben. Du kannst selbst vergleichen. Damit kann ich etwas anfangen. Und die kleine rote Lampe bleibt bitte an. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/fdff19b34fb020b6f72bddbf71477fcaeb6beeb288e31497727950f68bf630c6.mp3",
    "review_candidates": [
      "de-conversation-02-base17-block-free-u00-c00"
    ]
  },
  {
    "id": "de-conversation-02-base17-block-free-n4",
    "case": "de-conversation-02",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 4,
    "block_size": 5,
    "plain": "Stell dir ein winziges Radiostudio vor, in dem zwei Leute dieses Modell entdecken. Bitte sag mir, dass dort eine völlig übertriebene rote Aufnahmelampe leuchtet. Natürlich. Humaneness Voice Large liefert die Grundlage, das Podcasttraining übt den Dialog. Darf die zweite Stimme auch skeptisch nachfragen, statt immer nur freundlich zuzustimmen? Unbedingt. Jede Äußerung bekommt ihren eigenen Text und eine eigene emotionale Richtung. Von einer begeisterten Beschreibung lasse ich mich noch lange nicht überzeugen. Ich will es hören. Deshalb stehen die Referenzclips direkt neben den Ausgaben. Du kannst selbst vergleichen. Damit kann ich etwas anfangen. Und die kleine rote Lampe bleibt bitte an. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/2ab0a9e3479fd0cc0c08a74dc416fa6769b05027783dcd721f8c5e254fefb895.mp3",
    "review_candidates": [
      "de-conversation-02-base17-block-free-u00-c00"
    ]
  },
  {
    "id": "de-conversation-02-base17-block-free-n8",
    "case": "de-conversation-02",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 8,
    "block_size": 5,
    "plain": "Stell dir ein winziges Radiostudio vor, in dem zwei Leute dieses Modell entdecken. Bitte sag mir, dass dort eine völlig übertriebene rote Aufnahmelampe leuchtet. Natürlich. Humaneness Voice Large liefert die Grundlage, das Podcasttraining übt den Dialog. Darf die zweite Stimme auch skeptisch nachfragen, statt immer nur freundlich zuzustimmen? Unbedingt. Jede Äußerung bekommt ihren eigenen Text und eine eigene emotionale Richtung. Von einer begeisterten Beschreibung lasse ich mich noch lange nicht überzeugen. Ich will es hören. Deshalb stehen die Referenzclips direkt neben den Ausgaben. Du kannst selbst vergleichen. Damit kann ich etwas anfangen. Und die kleine rote Lampe bleibt bitte an. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/75d8fccb1c5c5ab4a55b3677d0e3e2d179cdac5e4f034f0b61bdb4f6c3d1d2b6.mp3",
    "review_candidates": [
      "de-conversation-02-base17-block-free-u00-c00"
    ]
  },
  {
    "id": "de-conversation-02-base17-block-timed-n1",
    "case": "de-conversation-02",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 1,
    "block_size": 5,
    "plain": "Stell dir ein winziges Radiostudio vor, in dem zwei Leute dieses Modell entdecken. Bitte sag mir, dass dort eine völlig übertriebene rote Aufnahmelampe leuchtet. Natürlich. Humaneness Voice Large liefert die Grundlage, das Podcasttraining übt den Dialog. Darf die zweite Stimme auch skeptisch nachfragen, statt immer nur freundlich zuzustimmen? Unbedingt. Jede Äußerung bekommt ihren eigenen Text und eine eigene emotionale Richtung. Von einer begeisterten Beschreibung lasse ich mich noch lange nicht überzeugen. Ich will es hören. Deshalb stehen die Referenzclips direkt neben den Ausgaben. Du kannst selbst vergleichen. Damit kann ich etwas anfangen. Und die kleine rote Lampe bleibt bitte an. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/6925b5827037d33703ba3fde6629a9933858a641d33e25f862ad777f0513ddd9.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-02-base17-block-timed-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 1025280
      },
      {
        "candidate_id": "de-conversation-02-base17-block-timed-u01-c00",
        "output_offset_sample": 1040640,
        "first_word_output_sample": 1044480,
        "last_word_output_sample": 2173440
      },
      {
        "candidate_id": "de-conversation-02-base17-block-timed-u02-c00",
        "output_offset_sample": 2188800,
        "first_word_output_sample": 2192640,
        "last_word_output_sample": 2672640
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 55.83,
    "samples": 2679840,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-02-base17-block-timed-n1.mp3"
  },
  {
    "id": "de-conversation-02-base17-block-timed-n4",
    "case": "de-conversation-02",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 4,
    "block_size": 5,
    "plain": "Stell dir ein winziges Radiostudio vor, in dem zwei Leute dieses Modell entdecken. Bitte sag mir, dass dort eine völlig übertriebene rote Aufnahmelampe leuchtet. Natürlich. Humaneness Voice Large liefert die Grundlage, das Podcasttraining übt den Dialog. Darf die zweite Stimme auch skeptisch nachfragen, statt immer nur freundlich zuzustimmen? Unbedingt. Jede Äußerung bekommt ihren eigenen Text und eine eigene emotionale Richtung. Von einer begeisterten Beschreibung lasse ich mich noch lange nicht überzeugen. Ich will es hören. Deshalb stehen die Referenzclips direkt neben den Ausgaben. Du kannst selbst vergleichen. Damit kann ich etwas anfangen. Und die kleine rote Lampe bleibt bitte an. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/8003cd86a876cbc3a31db92546346fa22b86111759c42b376d660c0fd7a946f9.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-02-base17-block-timed-u00-c01",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 1029120
      },
      {
        "candidate_id": "de-conversation-02-base17-block-timed-u01-c01",
        "output_offset_sample": 1044480,
        "first_word_output_sample": 1048320,
        "last_word_output_sample": 2173440
      },
      {
        "candidate_id": "de-conversation-02-base17-block-timed-u02-c01",
        "output_offset_sample": 2192640,
        "first_word_output_sample": 2192640,
        "last_word_output_sample": 2680320
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 55.99,
    "samples": 2687520,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-02-base17-block-timed-n4.mp3"
  },
  {
    "id": "de-conversation-02-base17-block-timed-n8",
    "case": "de-conversation-02",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 8,
    "block_size": 5,
    "plain": "Stell dir ein winziges Radiostudio vor, in dem zwei Leute dieses Modell entdecken. Bitte sag mir, dass dort eine völlig übertriebene rote Aufnahmelampe leuchtet. Natürlich. Humaneness Voice Large liefert die Grundlage, das Podcasttraining übt den Dialog. Darf die zweite Stimme auch skeptisch nachfragen, statt immer nur freundlich zuzustimmen? Unbedingt. Jede Äußerung bekommt ihren eigenen Text und eine eigene emotionale Richtung. Von einer begeisterten Beschreibung lasse ich mich noch lange nicht überzeugen. Ich will es hören. Deshalb stehen die Referenzclips direkt neben den Ausgaben. Du kannst selbst vergleichen. Damit kann ich etwas anfangen. Und die kleine rote Lampe bleibt bitte an. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/8003cd86a876cbc3a31db92546346fa22b86111759c42b376d660c0fd7a946f9.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-02-base17-block-timed-u00-c01",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 1029120
      },
      {
        "candidate_id": "de-conversation-02-base17-block-timed-u01-c01",
        "output_offset_sample": 1044480,
        "first_word_output_sample": 1048320,
        "last_word_output_sample": 2173440
      },
      {
        "candidate_id": "de-conversation-02-base17-block-timed-u02-c01",
        "output_offset_sample": 2192640,
        "first_word_output_sample": 2192640,
        "last_word_output_sample": 2680320
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 55.99,
    "samples": 2687520,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-02-base17-block-timed-n8.mp3"
  },
  {
    "id": "de-conversation-02-base17-segment-free-n1",
    "case": "de-conversation-02",
    "model": "base17",
    "strategy": "segment",
    "timing": "free",
    "best_of": 1,
    "block_size": 1,
    "plain": "Stell dir ein winziges Radiostudio vor, in dem zwei Leute dieses Modell entdecken. Bitte sag mir, dass dort eine völlig übertriebene rote Aufnahmelampe leuchtet. Natürlich. Humaneness Voice Large liefert die Grundlage, das Podcasttraining übt den Dialog. Darf die zweite Stimme auch skeptisch nachfragen, statt immer nur freundlich zuzustimmen? Unbedingt. Jede Äußerung bekommt ihren eigenen Text und eine eigene emotionale Richtung. Von einer begeisterten Beschreibung lasse ich mich noch lange nicht überzeugen. Ich will es hören. Deshalb stehen die Referenzclips direkt neben den Ausgaben. Du kannst selbst vergleichen. Damit kann ich etwas anfangen. Und die kleine rote Lampe bleibt bitte an. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": null,
    "review_candidates": [
      "de-conversation-02-base17-segment-free-u00-c00"
    ]
  },
  {
    "id": "de-conversation-02-base17-segment-timed-n1",
    "case": "de-conversation-02",
    "model": "base17",
    "strategy": "segment",
    "timing": "timed",
    "best_of": 1,
    "block_size": 1,
    "plain": "Stell dir ein winziges Radiostudio vor, in dem zwei Leute dieses Modell entdecken. Bitte sag mir, dass dort eine völlig übertriebene rote Aufnahmelampe leuchtet. Natürlich. Humaneness Voice Large liefert die Grundlage, das Podcasttraining übt den Dialog. Darf die zweite Stimme auch skeptisch nachfragen, statt immer nur freundlich zuzustimmen? Unbedingt. Jede Äußerung bekommt ihren eigenen Text und eine eigene emotionale Richtung. Von einer begeisterten Beschreibung lasse ich mich noch lange nicht überzeugen. Ich will es hören. Deshalb stehen die Referenzclips direkt neben den Ausgaben. Du kannst selbst vergleichen. Damit kann ich etwas anfangen. Und die kleine rote Lampe bleibt bitte an. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/f3198f1fe3a7b8b589066b1356857b60741355fbdfd8961e90536af739cfcfba.mp3",
    "review_candidates": [
      "de-conversation-02-base17-segment-timed-u01-c00",
      "de-conversation-02-base17-segment-timed-u04-c00"
    ]
  },
  {
    "id": "de-conversation-02-podcast-block-free-n1",
    "case": "de-conversation-02",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 1,
    "block_size": 5,
    "plain": "Stell dir ein winziges Radiostudio vor, in dem zwei Leute dieses Modell entdecken. Bitte sag mir, dass dort eine völlig übertriebene rote Aufnahmelampe leuchtet. Natürlich. Humaneness Voice Large liefert die Grundlage, das Podcasttraining übt den Dialog. Darf die zweite Stimme auch skeptisch nachfragen, statt immer nur freundlich zuzustimmen? Unbedingt. Jede Äußerung bekommt ihren eigenen Text und eine eigene emotionale Richtung. Von einer begeisterten Beschreibung lasse ich mich noch lange nicht überzeugen. Ich will es hören. Deshalb stehen die Referenzclips direkt neben den Ausgaben. Du kannst selbst vergleichen. Damit kann ich etwas anfangen. Und die kleine rote Lampe bleibt bitte an. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/12cc0de661679be9369b1a12c031b777104d3ebcb1de670d836ebbee4dd973a8.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-02-podcast-block-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 1117440
      },
      {
        "candidate_id": "de-conversation-02-podcast-block-free-u01-c00",
        "output_offset_sample": 1136640,
        "first_word_output_sample": 1136640,
        "last_word_output_sample": 2257920
      },
      {
        "candidate_id": "de-conversation-02-podcast-block-free-u02-c00",
        "output_offset_sample": 2277120,
        "first_word_output_sample": 2277120,
        "last_word_output_sample": 2810880
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 58.71,
    "samples": 2818080,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-02-podcast-block-free-n1.mp3"
  },
  {
    "id": "de-conversation-02-podcast-block-free-n4",
    "case": "de-conversation-02",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 4,
    "block_size": 5,
    "plain": "Stell dir ein winziges Radiostudio vor, in dem zwei Leute dieses Modell entdecken. Bitte sag mir, dass dort eine völlig übertriebene rote Aufnahmelampe leuchtet. Natürlich. Humaneness Voice Large liefert die Grundlage, das Podcasttraining übt den Dialog. Darf die zweite Stimme auch skeptisch nachfragen, statt immer nur freundlich zuzustimmen? Unbedingt. Jede Äußerung bekommt ihren eigenen Text und eine eigene emotionale Richtung. Von einer begeisterten Beschreibung lasse ich mich noch lange nicht überzeugen. Ich will es hören. Deshalb stehen die Referenzclips direkt neben den Ausgaben. Du kannst selbst vergleichen. Damit kann ich etwas anfangen. Und die kleine rote Lampe bleibt bitte an. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/39dcd65781939cd1a6a091827952e66a41e302be8594119f028baf838f74e8ae.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-02-podcast-block-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 1117440
      },
      {
        "candidate_id": "de-conversation-02-podcast-block-free-u01-c03",
        "output_offset_sample": 1136640,
        "first_word_output_sample": 1136640,
        "last_word_output_sample": 2280960
      },
      {
        "candidate_id": "de-conversation-02-podcast-block-free-u02-c03",
        "output_offset_sample": 2300160,
        "first_word_output_sample": 2300160,
        "last_word_output_sample": 2833920
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 59.19,
    "samples": 2841120,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-02-podcast-block-free-n4.mp3"
  },
  {
    "id": "de-conversation-02-podcast-block-free-n8",
    "case": "de-conversation-02",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 8,
    "block_size": 5,
    "plain": "Stell dir ein winziges Radiostudio vor, in dem zwei Leute dieses Modell entdecken. Bitte sag mir, dass dort eine völlig übertriebene rote Aufnahmelampe leuchtet. Natürlich. Humaneness Voice Large liefert die Grundlage, das Podcasttraining übt den Dialog. Darf die zweite Stimme auch skeptisch nachfragen, statt immer nur freundlich zuzustimmen? Unbedingt. Jede Äußerung bekommt ihren eigenen Text und eine eigene emotionale Richtung. Von einer begeisterten Beschreibung lasse ich mich noch lange nicht überzeugen. Ich will es hören. Deshalb stehen die Referenzclips direkt neben den Ausgaben. Du kannst selbst vergleichen. Damit kann ich etwas anfangen. Und die kleine rote Lampe bleibt bitte an. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/45f181bed4a1a3bfd20bfc98c5f571ad2708f9771ef9bb20d804c8585612421b.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-02-podcast-block-free-u00-c05",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 1121280
      },
      {
        "candidate_id": "de-conversation-02-podcast-block-free-u01-c03",
        "output_offset_sample": 1140480,
        "first_word_output_sample": 1140480,
        "last_word_output_sample": 2284800
      },
      {
        "candidate_id": "de-conversation-02-podcast-block-free-u02-c06",
        "output_offset_sample": 2304000,
        "first_word_output_sample": 2304000,
        "last_word_output_sample": 2826240
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 59.03,
    "samples": 2833440,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-02-podcast-block-free-n8.mp3"
  },
  {
    "id": "de-conversation-02-podcast-block-timed-n1",
    "case": "de-conversation-02",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 1,
    "block_size": 5,
    "plain": "Stell dir ein winziges Radiostudio vor, in dem zwei Leute dieses Modell entdecken. Bitte sag mir, dass dort eine völlig übertriebene rote Aufnahmelampe leuchtet. Natürlich. Humaneness Voice Large liefert die Grundlage, das Podcasttraining übt den Dialog. Darf die zweite Stimme auch skeptisch nachfragen, statt immer nur freundlich zuzustimmen? Unbedingt. Jede Äußerung bekommt ihren eigenen Text und eine eigene emotionale Richtung. Von einer begeisterten Beschreibung lasse ich mich noch lange nicht überzeugen. Ich will es hören. Deshalb stehen die Referenzclips direkt neben den Ausgaben. Du kannst selbst vergleichen. Damit kann ich etwas anfangen. Und die kleine rote Lampe bleibt bitte an. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/63060db3d996a9178bead027a45bc064f82619af2d963c51cad958778f7f47ad.mp3",
    "review_candidates": [
      "de-conversation-02-podcast-block-timed-u00-c00",
      "de-conversation-02-podcast-block-timed-u01-c00"
    ]
  },
  {
    "id": "de-conversation-02-podcast-block-timed-n4",
    "case": "de-conversation-02",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 4,
    "block_size": 5,
    "plain": "Stell dir ein winziges Radiostudio vor, in dem zwei Leute dieses Modell entdecken. Bitte sag mir, dass dort eine völlig übertriebene rote Aufnahmelampe leuchtet. Natürlich. Humaneness Voice Large liefert die Grundlage, das Podcasttraining übt den Dialog. Darf die zweite Stimme auch skeptisch nachfragen, statt immer nur freundlich zuzustimmen? Unbedingt. Jede Äußerung bekommt ihren eigenen Text und eine eigene emotionale Richtung. Von einer begeisterten Beschreibung lasse ich mich noch lange nicht überzeugen. Ich will es hören. Deshalb stehen die Referenzclips direkt neben den Ausgaben. Du kannst selbst vergleichen. Damit kann ich etwas anfangen. Und die kleine rote Lampe bleibt bitte an. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/9d059d14c23fd3532c67b2cf4be29196690d4d08ef322c63ffd63158f240a23f.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-02-podcast-block-timed-u00-c01",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 1029120
      },
      {
        "candidate_id": "de-conversation-02-podcast-block-timed-u01-c01",
        "output_offset_sample": 1048320,
        "first_word_output_sample": 1048320,
        "last_word_output_sample": 2188800
      },
      {
        "candidate_id": "de-conversation-02-podcast-block-timed-u02-c02",
        "output_offset_sample": 2208000,
        "first_word_output_sample": 2208000,
        "last_word_output_sample": 2699520
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 56.39,
    "samples": 2706720,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-02-podcast-block-timed-n4.mp3"
  },
  {
    "id": "de-conversation-02-podcast-block-timed-n8",
    "case": "de-conversation-02",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 8,
    "block_size": 5,
    "plain": "Stell dir ein winziges Radiostudio vor, in dem zwei Leute dieses Modell entdecken. Bitte sag mir, dass dort eine völlig übertriebene rote Aufnahmelampe leuchtet. Natürlich. Humaneness Voice Large liefert die Grundlage, das Podcasttraining übt den Dialog. Darf die zweite Stimme auch skeptisch nachfragen, statt immer nur freundlich zuzustimmen? Unbedingt. Jede Äußerung bekommt ihren eigenen Text und eine eigene emotionale Richtung. Von einer begeisterten Beschreibung lasse ich mich noch lange nicht überzeugen. Ich will es hören. Deshalb stehen die Referenzclips direkt neben den Ausgaben. Du kannst selbst vergleichen. Damit kann ich etwas anfangen. Und die kleine rote Lampe bleibt bitte an. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/9d059d14c23fd3532c67b2cf4be29196690d4d08ef322c63ffd63158f240a23f.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-02-podcast-block-timed-u00-c01",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 1029120
      },
      {
        "candidate_id": "de-conversation-02-podcast-block-timed-u01-c01",
        "output_offset_sample": 1048320,
        "first_word_output_sample": 1048320,
        "last_word_output_sample": 2188800
      },
      {
        "candidate_id": "de-conversation-02-podcast-block-timed-u02-c02",
        "output_offset_sample": 2208000,
        "first_word_output_sample": 2208000,
        "last_word_output_sample": 2699520
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 56.39,
    "samples": 2706720,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-02-podcast-block-timed-n8.mp3"
  },
  {
    "id": "de-conversation-02-podcast-segment-free-n1",
    "case": "de-conversation-02",
    "model": "podcast",
    "strategy": "segment",
    "timing": "free",
    "best_of": 1,
    "block_size": 1,
    "plain": "Stell dir ein winziges Radiostudio vor, in dem zwei Leute dieses Modell entdecken. Bitte sag mir, dass dort eine völlig übertriebene rote Aufnahmelampe leuchtet. Natürlich. Humaneness Voice Large liefert die Grundlage, das Podcasttraining übt den Dialog. Darf die zweite Stimme auch skeptisch nachfragen, statt immer nur freundlich zuzustimmen? Unbedingt. Jede Äußerung bekommt ihren eigenen Text und eine eigene emotionale Richtung. Von einer begeisterten Beschreibung lasse ich mich noch lange nicht überzeugen. Ich will es hören. Deshalb stehen die Referenzclips direkt neben den Ausgaben. Du kannst selbst vergleichen. Damit kann ich etwas anfangen. Und die kleine rote Lampe bleibt bitte an. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/f89160edf7f448a7bdbb625dbcf7ef8aeda0d65b014a74827b36fa7c8120dbfa.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-02-podcast-segment-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 280320
      },
      {
        "candidate_id": "de-conversation-02-podcast-segment-free-u01-c00",
        "output_offset_sample": 295680,
        "first_word_output_sample": 299520,
        "last_word_output_sample": 545280
      },
      {
        "candidate_id": "de-conversation-02-podcast-segment-free-u02-c00",
        "output_offset_sample": 560640,
        "first_word_output_sample": 564480,
        "last_word_output_sample": 867840
      },
      {
        "candidate_id": "de-conversation-02-podcast-segment-free-u03-c00",
        "output_offset_sample": 883200,
        "first_word_output_sample": 887040,
        "last_word_output_sample": 1171200
      },
      {
        "candidate_id": "de-conversation-02-podcast-segment-free-u04-c00",
        "output_offset_sample": 1190400,
        "first_word_output_sample": 1190400,
        "last_word_output_sample": 1482240
      },
      {
        "candidate_id": "de-conversation-02-podcast-segment-free-u05-c00",
        "output_offset_sample": 1501440,
        "first_word_output_sample": 1501440,
        "last_word_output_sample": 1777920
      },
      {
        "candidate_id": "de-conversation-02-podcast-segment-free-u06-c00",
        "output_offset_sample": 1797120,
        "first_word_output_sample": 1797120,
        "last_word_output_sample": 2138880
      },
      {
        "candidate_id": "de-conversation-02-podcast-segment-free-u07-c00",
        "output_offset_sample": 2154240,
        "first_word_output_sample": 2158080,
        "last_word_output_sample": 2365440
      },
      {
        "candidate_id": "de-conversation-02-podcast-segment-free-u08-c00",
        "output_offset_sample": 2380800,
        "first_word_output_sample": 2384640,
        "last_word_output_sample": 2684160
      },
      {
        "candidate_id": "de-conversation-02-podcast-segment-free-u09-c00",
        "output_offset_sample": 2703360,
        "first_word_output_sample": 2703360,
        "last_word_output_sample": 3018240
      },
      {
        "candidate_id": "de-conversation-02-podcast-segment-free-u10-c00",
        "output_offset_sample": 3037440,
        "first_word_output_sample": 3037440,
        "last_word_output_sample": 3429120
      },
      {
        "candidate_id": "de-conversation-02-podcast-segment-free-u11-c00",
        "output_offset_sample": 3444480,
        "first_word_output_sample": 3448320,
        "last_word_output_sample": 3747840
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 78.23,
    "samples": 3755040,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-02-podcast-segment-free-n1.mp3"
  },
  {
    "id": "de-conversation-02-podcast-segment-timed-n1",
    "case": "de-conversation-02",
    "model": "podcast",
    "strategy": "segment",
    "timing": "timed",
    "best_of": 1,
    "block_size": 1,
    "plain": "Stell dir ein winziges Radiostudio vor, in dem zwei Leute dieses Modell entdecken. Bitte sag mir, dass dort eine völlig übertriebene rote Aufnahmelampe leuchtet. Natürlich. Humaneness Voice Large liefert die Grundlage, das Podcasttraining übt den Dialog. Darf die zweite Stimme auch skeptisch nachfragen, statt immer nur freundlich zuzustimmen? Unbedingt. Jede Äußerung bekommt ihren eigenen Text und eine eigene emotionale Richtung. Von einer begeisterten Beschreibung lasse ich mich noch lange nicht überzeugen. Ich will es hören. Deshalb stehen die Referenzclips direkt neben den Ausgaben. Du kannst selbst vergleichen. Damit kann ich etwas anfangen. Und die kleine rote Lampe bleibt bitte an. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/1cd08f6940024f5434bb26c04b65acc08b023f6cc3a95ebc129eebb5f680ca7b.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-02-podcast-segment-timed-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 226560
      },
      {
        "candidate_id": "de-conversation-02-podcast-segment-timed-u01-c00",
        "output_offset_sample": 245760,
        "first_word_output_sample": 245760,
        "last_word_output_sample": 437760
      },
      {
        "candidate_id": "de-conversation-02-podcast-segment-timed-u02-c00",
        "output_offset_sample": 453120,
        "first_word_output_sample": 456960,
        "last_word_output_sample": 660480
      },
      {
        "candidate_id": "de-conversation-02-podcast-segment-timed-u03-c00",
        "output_offset_sample": 679680,
        "first_word_output_sample": 679680,
        "last_word_output_sample": 890880
      },
      {
        "candidate_id": "de-conversation-02-podcast-segment-timed-u04-c00",
        "output_offset_sample": 906240,
        "first_word_output_sample": 910080,
        "last_word_output_sample": 1159680
      },
      {
        "candidate_id": "de-conversation-02-podcast-segment-timed-u05-c00",
        "output_offset_sample": 1178880,
        "first_word_output_sample": 1178880,
        "last_word_output_sample": 1428480
      },
      {
        "candidate_id": "de-conversation-02-podcast-segment-timed-u06-c00",
        "output_offset_sample": 1447680,
        "first_word_output_sample": 1447680,
        "last_word_output_sample": 1655040
      },
      {
        "candidate_id": "de-conversation-02-podcast-segment-timed-u07-c00",
        "output_offset_sample": 1670400,
        "first_word_output_sample": 1674240,
        "last_word_output_sample": 1893120
      },
      {
        "candidate_id": "de-conversation-02-podcast-segment-timed-u08-c00",
        "output_offset_sample": 1912320,
        "first_word_output_sample": 1912320,
        "last_word_output_sample": 2138880
      },
      {
        "candidate_id": "de-conversation-02-podcast-segment-timed-u09-c00",
        "output_offset_sample": 2158080,
        "first_word_output_sample": 2158080,
        "last_word_output_sample": 2392320
      },
      {
        "candidate_id": "de-conversation-02-podcast-segment-timed-u10-c00",
        "output_offset_sample": 2411520,
        "first_word_output_sample": 2411520,
        "last_word_output_sample": 2680320
      },
      {
        "candidate_id": "de-conversation-02-podcast-segment-timed-u11-c00",
        "output_offset_sample": 2695680,
        "first_word_output_sample": 2699520,
        "last_word_output_sample": 2914560
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 60.87,
    "samples": 2921760,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-02-podcast-segment-timed-n1.mp3"
  },
  {
    "id": "de-conversation-03-base17-block-free-n1",
    "case": "de-conversation-03",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 1,
    "block_size": 4,
    "plain": "Mich interessiert besonders, ob kurze Aufnahmen die Identität beider Stimmen festlegen können. Und ob diese Stimmen erkennbar bleiben, wenn das Gespräch eine unerwartete Wendung nimmt? Ja. Das Modell erhält markierte Sprecherreferenzen und manchmal auch den bisherigen Gesprächsverlauf. Das wäre toll, aber gerade bei langen Gesprächen können Stimmen allmählich abdriften. Darum vergleichen wir die einzelnen Turns und vertrauen nicht einfach dem ersten Satz. Gut. Die begeisterte Gastgeberin soll dieselbe Person bleiben, wenn sie plötzlich zweifelt. Referenzen steuern die Identität, Spielanweisungen den Ausdruck. Beides ist keine Zaubergarantie. Seltsam, die ehrliche Erklärung macht mich neugieriger als das große Werbeversprechen. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/133e672a4cf16a44c24801ccf84223aac47108fa89ab195e7297ebce5394d2b3.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-03-base17-block-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 913920
      },
      {
        "candidate_id": "de-conversation-03-base17-block-free-u01-c00",
        "output_offset_sample": 929280,
        "first_word_output_sample": 933120,
        "last_word_output_sample": 1877760
      },
      {
        "candidate_id": "de-conversation-03-base17-block-free-u02-c00",
        "output_offset_sample": 1896960,
        "first_word_output_sample": 1896960,
        "last_word_output_sample": 2818560
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 58.87,
    "samples": 2825760,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-03-base17-block-free-n1.mp3"
  },
  {
    "id": "de-conversation-03-base17-block-free-n4",
    "case": "de-conversation-03",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 4,
    "block_size": 4,
    "plain": "Mich interessiert besonders, ob kurze Aufnahmen die Identität beider Stimmen festlegen können. Und ob diese Stimmen erkennbar bleiben, wenn das Gespräch eine unerwartete Wendung nimmt? Ja. Das Modell erhält markierte Sprecherreferenzen und manchmal auch den bisherigen Gesprächsverlauf. Das wäre toll, aber gerade bei langen Gesprächen können Stimmen allmählich abdriften. Darum vergleichen wir die einzelnen Turns und vertrauen nicht einfach dem ersten Satz. Gut. Die begeisterte Gastgeberin soll dieselbe Person bleiben, wenn sie plötzlich zweifelt. Referenzen steuern die Identität, Spielanweisungen den Ausdruck. Beides ist keine Zaubergarantie. Seltsam, die ehrliche Erklärung macht mich neugieriger als das große Werbeversprechen. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/2f918c6ce5aa3c60e7707a8058430f1466d7863291c9640cf704cd2e846dfe68.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-03-base17-block-free-u00-c03",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 925440
      },
      {
        "candidate_id": "de-conversation-03-base17-block-free-u01-c00",
        "output_offset_sample": 940800,
        "first_word_output_sample": 944640,
        "last_word_output_sample": 1889280
      },
      {
        "candidate_id": "de-conversation-03-base17-block-free-u02-c01",
        "output_offset_sample": 1904640,
        "first_word_output_sample": 1908480,
        "last_word_output_sample": 3041280
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 63.51,
    "samples": 3048480,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-03-base17-block-free-n4.mp3"
  },
  {
    "id": "de-conversation-03-base17-block-free-n8",
    "case": "de-conversation-03",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 8,
    "block_size": 4,
    "plain": "Mich interessiert besonders, ob kurze Aufnahmen die Identität beider Stimmen festlegen können. Und ob diese Stimmen erkennbar bleiben, wenn das Gespräch eine unerwartete Wendung nimmt? Ja. Das Modell erhält markierte Sprecherreferenzen und manchmal auch den bisherigen Gesprächsverlauf. Das wäre toll, aber gerade bei langen Gesprächen können Stimmen allmählich abdriften. Darum vergleichen wir die einzelnen Turns und vertrauen nicht einfach dem ersten Satz. Gut. Die begeisterte Gastgeberin soll dieselbe Person bleiben, wenn sie plötzlich zweifelt. Referenzen steuern die Identität, Spielanweisungen den Ausdruck. Beides ist keine Zaubergarantie. Seltsam, die ehrliche Erklärung macht mich neugieriger als das große Werbeversprechen. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/0f9104e6cfd8b2d41f9bcf2055364c802cb634931aa1d96da806fda6e30f4aab.mp3",
    "review_candidates": [
      "de-conversation-03-base17-block-free-u01-c05"
    ]
  },
  {
    "id": "de-conversation-03-base17-block-timed-n1",
    "case": "de-conversation-03",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 1,
    "block_size": 4,
    "plain": "Mich interessiert besonders, ob kurze Aufnahmen die Identität beider Stimmen festlegen können. Und ob diese Stimmen erkennbar bleiben, wenn das Gespräch eine unerwartete Wendung nimmt? Ja. Das Modell erhält markierte Sprecherreferenzen und manchmal auch den bisherigen Gesprächsverlauf. Das wäre toll, aber gerade bei langen Gesprächen können Stimmen allmählich abdriften. Darum vergleichen wir die einzelnen Turns und vertrauen nicht einfach dem ersten Satz. Gut. Die begeisterte Gastgeberin soll dieselbe Person bleiben, wenn sie plötzlich zweifelt. Referenzen steuern die Identität, Spielanweisungen den Ausdruck. Beides ist keine Zaubergarantie. Seltsam, die ehrliche Erklärung macht mich neugieriger als das große Werbeversprechen. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/de6e303052a0ad57ccca2b3888ca5010488ee0abe658c7cd72135cbf6c954c00.mp3",
    "review_candidates": [
      "de-conversation-03-base17-block-timed-u01-c00"
    ]
  },
  {
    "id": "de-conversation-03-base17-block-timed-n4",
    "case": "de-conversation-03",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 4,
    "block_size": 4,
    "plain": "Mich interessiert besonders, ob kurze Aufnahmen die Identität beider Stimmen festlegen können. Und ob diese Stimmen erkennbar bleiben, wenn das Gespräch eine unerwartete Wendung nimmt? Ja. Das Modell erhält markierte Sprecherreferenzen und manchmal auch den bisherigen Gesprächsverlauf. Das wäre toll, aber gerade bei langen Gesprächen können Stimmen allmählich abdriften. Darum vergleichen wir die einzelnen Turns und vertrauen nicht einfach dem ersten Satz. Gut. Die begeisterte Gastgeberin soll dieselbe Person bleiben, wenn sie plötzlich zweifelt. Referenzen steuern die Identität, Spielanweisungen den Ausdruck. Beides ist keine Zaubergarantie. Seltsam, die ehrliche Erklärung macht mich neugieriger als das große Werbeversprechen. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/963bfe5e681060c8eb0ff5b48332cee9f1210101e8295fa734da9fef23a7f9ed.mp3",
    "review_candidates": [
      "de-conversation-03-base17-block-timed-u01-c00"
    ]
  },
  {
    "id": "de-conversation-03-base17-block-timed-n8",
    "case": "de-conversation-03",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 8,
    "block_size": 4,
    "plain": "Mich interessiert besonders, ob kurze Aufnahmen die Identität beider Stimmen festlegen können. Und ob diese Stimmen erkennbar bleiben, wenn das Gespräch eine unerwartete Wendung nimmt? Ja. Das Modell erhält markierte Sprecherreferenzen und manchmal auch den bisherigen Gesprächsverlauf. Das wäre toll, aber gerade bei langen Gesprächen können Stimmen allmählich abdriften. Darum vergleichen wir die einzelnen Turns und vertrauen nicht einfach dem ersten Satz. Gut. Die begeisterte Gastgeberin soll dieselbe Person bleiben, wenn sie plötzlich zweifelt. Referenzen steuern die Identität, Spielanweisungen den Ausdruck. Beides ist keine Zaubergarantie. Seltsam, die ehrliche Erklärung macht mich neugieriger als das große Werbeversprechen. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/c696f8e90d33748c4c1a45bcab8e7d5a30bcd5f545921896ff76434824bb4066.mp3",
    "review_candidates": [
      "de-conversation-03-base17-block-timed-u01-c00"
    ]
  },
  {
    "id": "de-conversation-03-base17-segment-free-n1",
    "case": "de-conversation-03",
    "model": "base17",
    "strategy": "segment",
    "timing": "free",
    "best_of": 1,
    "block_size": 1,
    "plain": "Mich interessiert besonders, ob kurze Aufnahmen die Identität beider Stimmen festlegen können. Und ob diese Stimmen erkennbar bleiben, wenn das Gespräch eine unerwartete Wendung nimmt? Ja. Das Modell erhält markierte Sprecherreferenzen und manchmal auch den bisherigen Gesprächsverlauf. Das wäre toll, aber gerade bei langen Gesprächen können Stimmen allmählich abdriften. Darum vergleichen wir die einzelnen Turns und vertrauen nicht einfach dem ersten Satz. Gut. Die begeisterte Gastgeberin soll dieselbe Person bleiben, wenn sie plötzlich zweifelt. Referenzen steuern die Identität, Spielanweisungen den Ausdruck. Beides ist keine Zaubergarantie. Seltsam, die ehrliche Erklärung macht mich neugieriger als das große Werbeversprechen. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/ca1559eb9a9242f98308ce1eab152285dcf40b7dac09639e7a9f03ab8565444a.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-03-base17-segment-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 556800
      },
      {
        "candidate_id": "de-conversation-03-base17-segment-free-u01-c00",
        "output_offset_sample": 576000,
        "first_word_output_sample": 576000,
        "last_word_output_sample": 883200
      },
      {
        "candidate_id": "de-conversation-03-base17-segment-free-u02-c00",
        "output_offset_sample": 902400,
        "first_word_output_sample": 902400,
        "last_word_output_sample": 1240320
      },
      {
        "candidate_id": "de-conversation-03-base17-segment-free-u03-c00",
        "output_offset_sample": 1255680,
        "first_word_output_sample": 1259520,
        "last_word_output_sample": 1539840
      },
      {
        "candidate_id": "de-conversation-03-base17-segment-free-u04-c00",
        "output_offset_sample": 1555200,
        "first_word_output_sample": 1559040,
        "last_word_output_sample": 1847040
      },
      {
        "candidate_id": "de-conversation-03-base17-segment-free-u05-c00",
        "output_offset_sample": 1862400,
        "first_word_output_sample": 1866240,
        "last_word_output_sample": 2169600
      },
      {
        "candidate_id": "de-conversation-03-base17-segment-free-u06-c00",
        "output_offset_sample": 2184960,
        "first_word_output_sample": 2188800,
        "last_word_output_sample": 2568960
      },
      {
        "candidate_id": "de-conversation-03-base17-segment-free-u07-c00",
        "output_offset_sample": 2584320,
        "first_word_output_sample": 2588160,
        "last_word_output_sample": 2891520
      },
      {
        "candidate_id": "de-conversation-03-base17-segment-free-u08-c00",
        "output_offset_sample": 2910720,
        "first_word_output_sample": 2910720,
        "last_word_output_sample": 3244800
      },
      {
        "candidate_id": "de-conversation-03-base17-segment-free-u09-c00",
        "output_offset_sample": 3260160,
        "first_word_output_sample": 3264000,
        "last_word_output_sample": 3605760
      },
      {
        "candidate_id": "de-conversation-03-base17-segment-free-u10-c00",
        "output_offset_sample": 3621120,
        "first_word_output_sample": 3624960,
        "last_word_output_sample": 4243200
      },
      {
        "candidate_id": "de-conversation-03-base17-segment-free-u11-c00",
        "output_offset_sample": 4258560,
        "first_word_output_sample": 4262400,
        "last_word_output_sample": 4592640
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 95.83,
    "samples": 4599840,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-03-base17-segment-free-n1.mp3"
  },
  {
    "id": "de-conversation-03-base17-segment-timed-n1",
    "case": "de-conversation-03",
    "model": "base17",
    "strategy": "segment",
    "timing": "timed",
    "best_of": 1,
    "block_size": 1,
    "plain": "Mich interessiert besonders, ob kurze Aufnahmen die Identität beider Stimmen festlegen können. Und ob diese Stimmen erkennbar bleiben, wenn das Gespräch eine unerwartete Wendung nimmt? Ja. Das Modell erhält markierte Sprecherreferenzen und manchmal auch den bisherigen Gesprächsverlauf. Das wäre toll, aber gerade bei langen Gesprächen können Stimmen allmählich abdriften. Darum vergleichen wir die einzelnen Turns und vertrauen nicht einfach dem ersten Satz. Gut. Die begeisterte Gastgeberin soll dieselbe Person bleiben, wenn sie plötzlich zweifelt. Referenzen steuern die Identität, Spielanweisungen den Ausdruck. Beides ist keine Zaubergarantie. Seltsam, die ehrliche Erklärung macht mich neugieriger als das große Werbeversprechen. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/a87db38b7c9afac5e5d8354ea3dbd40879320896ae0803408fba86cb35494d98.mp3",
    "review_candidates": [
      "de-conversation-03-base17-segment-timed-u00-c00",
      "de-conversation-03-base17-segment-timed-u02-c00",
      "de-conversation-03-base17-segment-timed-u06-c00",
      "de-conversation-03-base17-segment-timed-u11-c00"
    ]
  },
  {
    "id": "de-conversation-03-podcast-block-free-n1",
    "case": "de-conversation-03",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 1,
    "block_size": 4,
    "plain": "Mich interessiert besonders, ob kurze Aufnahmen die Identität beider Stimmen festlegen können. Und ob diese Stimmen erkennbar bleiben, wenn das Gespräch eine unerwartete Wendung nimmt? Ja. Das Modell erhält markierte Sprecherreferenzen und manchmal auch den bisherigen Gesprächsverlauf. Das wäre toll, aber gerade bei langen Gesprächen können Stimmen allmählich abdriften. Darum vergleichen wir die einzelnen Turns und vertrauen nicht einfach dem ersten Satz. Gut. Die begeisterte Gastgeberin soll dieselbe Person bleiben, wenn sie plötzlich zweifelt. Referenzen steuern die Identität, Spielanweisungen den Ausdruck. Beides ist keine Zaubergarantie. Seltsam, die ehrliche Erklärung macht mich neugieriger als das große Werbeversprechen. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": null,
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-03-podcast-block-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 998400
      },
      {
        "candidate_id": "de-conversation-03-podcast-block-free-u01-c00",
        "output_offset_sample": 1017600,
        "first_word_output_sample": 1017600,
        "last_word_output_sample": 1992960
      },
      {
        "candidate_id": "de-conversation-03-podcast-block-free-u02-c00",
        "output_offset_sample": 2012160,
        "first_word_output_sample": 2012160,
        "last_word_output_sample": 2979840
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 62.23,
    "samples": 2987040,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-03-podcast-block-free-n1.mp3"
  },
  {
    "id": "de-conversation-03-podcast-block-free-n4",
    "case": "de-conversation-03",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 4,
    "block_size": 4,
    "plain": "Mich interessiert besonders, ob kurze Aufnahmen die Identität beider Stimmen festlegen können. Und ob diese Stimmen erkennbar bleiben, wenn das Gespräch eine unerwartete Wendung nimmt? Ja. Das Modell erhält markierte Sprecherreferenzen und manchmal auch den bisherigen Gesprächsverlauf. Das wäre toll, aber gerade bei langen Gesprächen können Stimmen allmählich abdriften. Darum vergleichen wir die einzelnen Turns und vertrauen nicht einfach dem ersten Satz. Gut. Die begeisterte Gastgeberin soll dieselbe Person bleiben, wenn sie plötzlich zweifelt. Referenzen steuern die Identität, Spielanweisungen den Ausdruck. Beides ist keine Zaubergarantie. Seltsam, die ehrliche Erklärung macht mich neugieriger als das große Werbeversprechen. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/4270d101ae43a618615ff4a0914b61d582e1f53d8dffbdfe5667896ec2077399.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-03-podcast-block-free-u00-c02",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 967680
      },
      {
        "candidate_id": "de-conversation-03-podcast-block-free-u01-c00",
        "output_offset_sample": 986880,
        "first_word_output_sample": 986880,
        "last_word_output_sample": 1962240
      },
      {
        "candidate_id": "de-conversation-03-podcast-block-free-u02-c00",
        "output_offset_sample": 1981440,
        "first_word_output_sample": 1981440,
        "last_word_output_sample": 2949120
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 61.59,
    "samples": 2956320,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-03-podcast-block-free-n4.mp3"
  },
  {
    "id": "de-conversation-03-podcast-block-free-n8",
    "case": "de-conversation-03",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 8,
    "block_size": 4,
    "plain": "Mich interessiert besonders, ob kurze Aufnahmen die Identität beider Stimmen festlegen können. Und ob diese Stimmen erkennbar bleiben, wenn das Gespräch eine unerwartete Wendung nimmt? Ja. Das Modell erhält markierte Sprecherreferenzen und manchmal auch den bisherigen Gesprächsverlauf. Das wäre toll, aber gerade bei langen Gesprächen können Stimmen allmählich abdriften. Darum vergleichen wir die einzelnen Turns und vertrauen nicht einfach dem ersten Satz. Gut. Die begeisterte Gastgeberin soll dieselbe Person bleiben, wenn sie plötzlich zweifelt. Referenzen steuern die Identität, Spielanweisungen den Ausdruck. Beides ist keine Zaubergarantie. Seltsam, die ehrliche Erklärung macht mich neugieriger als das große Werbeversprechen. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/4270d101ae43a618615ff4a0914b61d582e1f53d8dffbdfe5667896ec2077399.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-03-podcast-block-free-u00-c02",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 967680
      },
      {
        "candidate_id": "de-conversation-03-podcast-block-free-u01-c00",
        "output_offset_sample": 986880,
        "first_word_output_sample": 986880,
        "last_word_output_sample": 1962240
      },
      {
        "candidate_id": "de-conversation-03-podcast-block-free-u02-c00",
        "output_offset_sample": 1981440,
        "first_word_output_sample": 1981440,
        "last_word_output_sample": 2949120
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 61.59,
    "samples": 2956320,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-03-podcast-block-free-n8.mp3"
  },
  {
    "id": "de-conversation-03-podcast-block-timed-n1",
    "case": "de-conversation-03",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 1,
    "block_size": 4,
    "plain": "Mich interessiert besonders, ob kurze Aufnahmen die Identität beider Stimmen festlegen können. Und ob diese Stimmen erkennbar bleiben, wenn das Gespräch eine unerwartete Wendung nimmt? Ja. Das Modell erhält markierte Sprecherreferenzen und manchmal auch den bisherigen Gesprächsverlauf. Das wäre toll, aber gerade bei langen Gesprächen können Stimmen allmählich abdriften. Darum vergleichen wir die einzelnen Turns und vertrauen nicht einfach dem ersten Satz. Gut. Die begeisterte Gastgeberin soll dieselbe Person bleiben, wenn sie plötzlich zweifelt. Referenzen steuern die Identität, Spielanweisungen den Ausdruck. Beides ist keine Zaubergarantie. Seltsam, die ehrliche Erklärung macht mich neugieriger als das große Werbeversprechen. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/50bb126dab564c237358f28fcd115c54c6e66c553c237f6b1a4b0b15ed090f74.mp3",
    "review_candidates": [
      "de-conversation-03-podcast-block-timed-u01-c00"
    ]
  },
  {
    "id": "de-conversation-03-podcast-block-timed-n4",
    "case": "de-conversation-03",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 4,
    "block_size": 4,
    "plain": "Mich interessiert besonders, ob kurze Aufnahmen die Identität beider Stimmen festlegen können. Und ob diese Stimmen erkennbar bleiben, wenn das Gespräch eine unerwartete Wendung nimmt? Ja. Das Modell erhält markierte Sprecherreferenzen und manchmal auch den bisherigen Gesprächsverlauf. Das wäre toll, aber gerade bei langen Gesprächen können Stimmen allmählich abdriften. Darum vergleichen wir die einzelnen Turns und vertrauen nicht einfach dem ersten Satz. Gut. Die begeisterte Gastgeberin soll dieselbe Person bleiben, wenn sie plötzlich zweifelt. Referenzen steuern die Identität, Spielanweisungen den Ausdruck. Beides ist keine Zaubergarantie. Seltsam, die ehrliche Erklärung macht mich neugieriger als das große Werbeversprechen. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/65eea114912b910448abf243cb3f6fcb03f5e14e30f1c65c63a34583b59946fe.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-03-podcast-block-timed-u00-c03",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 837120
      },
      {
        "candidate_id": "de-conversation-03-podcast-block-timed-u01-c03",
        "output_offset_sample": 856320,
        "first_word_output_sample": 856320,
        "last_word_output_sample": 1655040
      },
      {
        "candidate_id": "de-conversation-03-podcast-block-timed-u02-c03",
        "output_offset_sample": 1674240,
        "first_word_output_sample": 1674240,
        "last_word_output_sample": 2634240
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 55.03,
    "samples": 2641440,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-03-podcast-block-timed-n4.mp3"
  },
  {
    "id": "de-conversation-03-podcast-block-timed-n8",
    "case": "de-conversation-03",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 8,
    "block_size": 4,
    "plain": "Mich interessiert besonders, ob kurze Aufnahmen die Identität beider Stimmen festlegen können. Und ob diese Stimmen erkennbar bleiben, wenn das Gespräch eine unerwartete Wendung nimmt? Ja. Das Modell erhält markierte Sprecherreferenzen und manchmal auch den bisherigen Gesprächsverlauf. Das wäre toll, aber gerade bei langen Gesprächen können Stimmen allmählich abdriften. Darum vergleichen wir die einzelnen Turns und vertrauen nicht einfach dem ersten Satz. Gut. Die begeisterte Gastgeberin soll dieselbe Person bleiben, wenn sie plötzlich zweifelt. Referenzen steuern die Identität, Spielanweisungen den Ausdruck. Beides ist keine Zaubergarantie. Seltsam, die ehrliche Erklärung macht mich neugieriger als das große Werbeversprechen. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/c0480c2d3a6a30ef5b453f2ba0c1dff1714cbc3f11b6cad1a49210971ceddac8.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-03-podcast-block-timed-u00-c03",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 837120
      },
      {
        "candidate_id": "de-conversation-03-podcast-block-timed-u01-c07",
        "output_offset_sample": 856320,
        "first_word_output_sample": 856320,
        "last_word_output_sample": 1655040
      },
      {
        "candidate_id": "de-conversation-03-podcast-block-timed-u02-c03",
        "output_offset_sample": 1674240,
        "first_word_output_sample": 1674240,
        "last_word_output_sample": 2634240
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 55.03,
    "samples": 2641440,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-03-podcast-block-timed-n8.mp3"
  },
  {
    "id": "de-conversation-03-podcast-segment-free-n1",
    "case": "de-conversation-03",
    "model": "podcast",
    "strategy": "segment",
    "timing": "free",
    "best_of": 1,
    "block_size": 1,
    "plain": "Mich interessiert besonders, ob kurze Aufnahmen die Identität beider Stimmen festlegen können. Und ob diese Stimmen erkennbar bleiben, wenn das Gespräch eine unerwartete Wendung nimmt? Ja. Das Modell erhält markierte Sprecherreferenzen und manchmal auch den bisherigen Gesprächsverlauf. Das wäre toll, aber gerade bei langen Gesprächen können Stimmen allmählich abdriften. Darum vergleichen wir die einzelnen Turns und vertrauen nicht einfach dem ersten Satz. Gut. Die begeisterte Gastgeberin soll dieselbe Person bleiben, wenn sie plötzlich zweifelt. Referenzen steuern die Identität, Spielanweisungen den Ausdruck. Beides ist keine Zaubergarantie. Seltsam, die ehrliche Erklärung macht mich neugieriger als das große Werbeversprechen. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/b27a2d2fb2b29754176c3fe92c797255ab9d78ad3657635ac3f3ecb90760a5dd.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-03-podcast-segment-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 303360
      },
      {
        "candidate_id": "de-conversation-03-podcast-segment-free-u01-c00",
        "output_offset_sample": 322560,
        "first_word_output_sample": 322560,
        "last_word_output_sample": 633600
      },
      {
        "candidate_id": "de-conversation-03-podcast-segment-free-u02-c00",
        "output_offset_sample": 648960,
        "first_word_output_sample": 652800,
        "last_word_output_sample": 948480
      },
      {
        "candidate_id": "de-conversation-03-podcast-segment-free-u03-c00",
        "output_offset_sample": 963840,
        "first_word_output_sample": 967680,
        "last_word_output_sample": 1251840
      },
      {
        "candidate_id": "de-conversation-03-podcast-segment-free-u04-c00",
        "output_offset_sample": 1267200,
        "first_word_output_sample": 1271040,
        "last_word_output_sample": 1524480
      },
      {
        "candidate_id": "de-conversation-03-podcast-segment-free-u05-c00",
        "output_offset_sample": 1539840,
        "first_word_output_sample": 1543680,
        "last_word_output_sample": 1873920
      },
      {
        "candidate_id": "de-conversation-03-podcast-segment-free-u06-c00",
        "output_offset_sample": 1889280,
        "first_word_output_sample": 1893120,
        "last_word_output_sample": 2200320
      },
      {
        "candidate_id": "de-conversation-03-podcast-segment-free-u07-c00",
        "output_offset_sample": 2215680,
        "first_word_output_sample": 2219520,
        "last_word_output_sample": 2499840
      },
      {
        "candidate_id": "de-conversation-03-podcast-segment-free-u08-c00",
        "output_offset_sample": 2519040,
        "first_word_output_sample": 2519040,
        "last_word_output_sample": 2818560
      },
      {
        "candidate_id": "de-conversation-03-podcast-segment-free-u09-c00",
        "output_offset_sample": 2837760,
        "first_word_output_sample": 2837760,
        "last_word_output_sample": 3175680
      },
      {
        "candidate_id": "de-conversation-03-podcast-segment-free-u10-c00",
        "output_offset_sample": 3191040,
        "first_word_output_sample": 3194880,
        "last_word_output_sample": 3502080
      },
      {
        "candidate_id": "de-conversation-03-podcast-segment-free-u11-c00",
        "output_offset_sample": 3517440,
        "first_word_output_sample": 3521280,
        "last_word_output_sample": 3832320
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 79.99,
    "samples": 3839520,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-03-podcast-segment-free-n1.mp3"
  },
  {
    "id": "de-conversation-03-podcast-segment-timed-n1",
    "case": "de-conversation-03",
    "model": "podcast",
    "strategy": "segment",
    "timing": "timed",
    "best_of": 1,
    "block_size": 1,
    "plain": "Mich interessiert besonders, ob kurze Aufnahmen die Identität beider Stimmen festlegen können. Und ob diese Stimmen erkennbar bleiben, wenn das Gespräch eine unerwartete Wendung nimmt? Ja. Das Modell erhält markierte Sprecherreferenzen und manchmal auch den bisherigen Gesprächsverlauf. Das wäre toll, aber gerade bei langen Gesprächen können Stimmen allmählich abdriften. Darum vergleichen wir die einzelnen Turns und vertrauen nicht einfach dem ersten Satz. Gut. Die begeisterte Gastgeberin soll dieselbe Person bleiben, wenn sie plötzlich zweifelt. Referenzen steuern die Identität, Spielanweisungen den Ausdruck. Beides ist keine Zaubergarantie. Seltsam, die ehrliche Erklärung macht mich neugieriger als das große Werbeversprechen. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/fd3332ed05bf0c2ccd33e0021b9bc388e368d4154c39c9cfe643698e89bae297.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-03-podcast-segment-timed-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 215040
      },
      {
        "candidate_id": "de-conversation-03-podcast-segment-timed-u01-c00",
        "output_offset_sample": 234240,
        "first_word_output_sample": 234240,
        "last_word_output_sample": 453120
      },
      {
        "candidate_id": "de-conversation-03-podcast-segment-timed-u02-c00",
        "output_offset_sample": 472320,
        "first_word_output_sample": 472320,
        "last_word_output_sample": 683520
      },
      {
        "candidate_id": "de-conversation-03-podcast-segment-timed-u03-c00",
        "output_offset_sample": 702720,
        "first_word_output_sample": 702720,
        "last_word_output_sample": 910080
      },
      {
        "candidate_id": "de-conversation-03-podcast-segment-timed-u04-c00",
        "output_offset_sample": 929280,
        "first_word_output_sample": 929280,
        "last_word_output_sample": 1148160
      },
      {
        "candidate_id": "de-conversation-03-podcast-segment-timed-u05-c00",
        "output_offset_sample": 1163520,
        "first_word_output_sample": 1167360,
        "last_word_output_sample": 1370880
      },
      {
        "candidate_id": "de-conversation-03-podcast-segment-timed-u06-c00",
        "output_offset_sample": 1390080,
        "first_word_output_sample": 1390080,
        "last_word_output_sample": 1593600
      },
      {
        "candidate_id": "de-conversation-03-podcast-segment-timed-u07-c00",
        "output_offset_sample": 1608960,
        "first_word_output_sample": 1612800,
        "last_word_output_sample": 1800960
      },
      {
        "candidate_id": "de-conversation-03-podcast-segment-timed-u08-c00",
        "output_offset_sample": 1820160,
        "first_word_output_sample": 1820160,
        "last_word_output_sample": 2046720
      },
      {
        "candidate_id": "de-conversation-03-podcast-segment-timed-u09-c00",
        "output_offset_sample": 2065920,
        "first_word_output_sample": 2065920,
        "last_word_output_sample": 2300160
      },
      {
        "candidate_id": "de-conversation-03-podcast-segment-timed-u10-c00",
        "output_offset_sample": 2315520,
        "first_word_output_sample": 2319360,
        "last_word_output_sample": 2588160
      },
      {
        "candidate_id": "de-conversation-03-podcast-segment-timed-u11-c00",
        "output_offset_sample": 2603520,
        "first_word_output_sample": 2607360,
        "last_word_output_sample": 2818560
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 58.87,
    "samples": 2825760,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-03-podcast-segment-timed-n1.mp3"
  },
  {
    "id": "de-conversation-04-base17-block-free-n1",
    "case": "de-conversation-04",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 1,
    "block_size": 5,
    "plain": "Im Skript kann ich neben einen einzelnen Satz eine gewünschte Dauer schreiben. Du machst Witze. Trifft das Modell dann wirklich jede einzelne Millisekunde? Nein. Der Codec arbeitet in Achtzig-Millisekunden-Schritten, und die gewünschte Länge kann verfehlt werden. Danke. Ich wollte gerade mein ganzes Leben auf drei Sekunden zusammenfassen lassen. Wir messen die tatsächliche Dauer und zeigen auch vorzeitige Enden und erreichte Längenlimits. Dann lass diesen Satz ruhig klingen, als würde ein Freund nach den richtigen Worten suchen. Genau darum geht es: eine Absicht vorgeben, das Ergebnis messen und aufmerksam zuhören. Ein angenehm menschlicher Umgang mit einer ziemlich eigenartigen kleinen Maschine. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/79e07cbb1958e74a0bcf256b978e0c41d89a9d4c3ae99466c9d57e55aa543821.mp3",
    "review_candidates": [
      "de-conversation-04-base17-block-free-u01-c00"
    ]
  },
  {
    "id": "de-conversation-04-base17-block-free-n4",
    "case": "de-conversation-04",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 4,
    "block_size": 5,
    "plain": "Im Skript kann ich neben einen einzelnen Satz eine gewünschte Dauer schreiben. Du machst Witze. Trifft das Modell dann wirklich jede einzelne Millisekunde? Nein. Der Codec arbeitet in Achtzig-Millisekunden-Schritten, und die gewünschte Länge kann verfehlt werden. Danke. Ich wollte gerade mein ganzes Leben auf drei Sekunden zusammenfassen lassen. Wir messen die tatsächliche Dauer und zeigen auch vorzeitige Enden und erreichte Längenlimits. Dann lass diesen Satz ruhig klingen, als würde ein Freund nach den richtigen Worten suchen. Genau darum geht es: eine Absicht vorgeben, das Ergebnis messen und aufmerksam zuhören. Ein angenehm menschlicher Umgang mit einer ziemlich eigenartigen kleinen Maschine. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/d9d0d311093111fa7f04f3d9ac15d7118ad2585741c942db90323e63e96ab918.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-04-base17-block-free-u00-c02",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 1065600
      },
      {
        "candidate_id": "de-conversation-04-base17-block-free-u01-c01",
        "output_offset_sample": 1080960,
        "first_word_output_sample": 1084800,
        "last_word_output_sample": 2144640
      },
      {
        "candidate_id": "de-conversation-04-base17-block-free-u02-c01",
        "output_offset_sample": 2163840,
        "first_word_output_sample": 2163840,
        "last_word_output_sample": 2749440
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 57.43,
    "samples": 2756640,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-04-base17-block-free-n4.mp3"
  },
  {
    "id": "de-conversation-04-base17-block-free-n8",
    "case": "de-conversation-04",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 8,
    "block_size": 5,
    "plain": "Im Skript kann ich neben einen einzelnen Satz eine gewünschte Dauer schreiben. Du machst Witze. Trifft das Modell dann wirklich jede einzelne Millisekunde? Nein. Der Codec arbeitet in Achtzig-Millisekunden-Schritten, und die gewünschte Länge kann verfehlt werden. Danke. Ich wollte gerade mein ganzes Leben auf drei Sekunden zusammenfassen lassen. Wir messen die tatsächliche Dauer und zeigen auch vorzeitige Enden und erreichte Längenlimits. Dann lass diesen Satz ruhig klingen, als würde ein Freund nach den richtigen Worten suchen. Genau darum geht es: eine Absicht vorgeben, das Ergebnis messen und aufmerksam zuhören. Ein angenehm menschlicher Umgang mit einer ziemlich eigenartigen kleinen Maschine. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/d9d0d311093111fa7f04f3d9ac15d7118ad2585741c942db90323e63e96ab918.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-04-base17-block-free-u00-c02",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 1065600
      },
      {
        "candidate_id": "de-conversation-04-base17-block-free-u01-c01",
        "output_offset_sample": 1080960,
        "first_word_output_sample": 1084800,
        "last_word_output_sample": 2144640
      },
      {
        "candidate_id": "de-conversation-04-base17-block-free-u02-c01",
        "output_offset_sample": 2163840,
        "first_word_output_sample": 2163840,
        "last_word_output_sample": 2749440
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 57.43,
    "samples": 2756640,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-04-base17-block-free-n8.mp3"
  },
  {
    "id": "de-conversation-04-base17-block-timed-n1",
    "case": "de-conversation-04",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 1,
    "block_size": 5,
    "plain": "Im Skript kann ich neben einen einzelnen Satz eine gewünschte Dauer schreiben. Du machst Witze. Trifft das Modell dann wirklich jede einzelne Millisekunde? Nein. Der Codec arbeitet in Achtzig-Millisekunden-Schritten, und die gewünschte Länge kann verfehlt werden. Danke. Ich wollte gerade mein ganzes Leben auf drei Sekunden zusammenfassen lassen. Wir messen die tatsächliche Dauer und zeigen auch vorzeitige Enden und erreichte Längenlimits. Dann lass diesen Satz ruhig klingen, als würde ein Freund nach den richtigen Worten suchen. Genau darum geht es: eine Absicht vorgeben, das Ergebnis messen und aufmerksam zuhören. Ein angenehm menschlicher Umgang mit einer ziemlich eigenartigen kleinen Maschine. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/6eb55ccf3dafc897d29198f37b07932b5c9b97653d8e0653ac5ea32a12d32ff1.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-04-base17-block-timed-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 1044480
      },
      {
        "candidate_id": "de-conversation-04-base17-block-timed-u01-c00",
        "output_offset_sample": 1059840,
        "first_word_output_sample": 1063680,
        "last_word_output_sample": 2154240
      },
      {
        "candidate_id": "de-conversation-04-base17-block-timed-u02-c00",
        "output_offset_sample": 2173440,
        "first_word_output_sample": 2173440,
        "last_word_output_sample": 2664960
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 55.67,
    "samples": 2672160,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-04-base17-block-timed-n1.mp3"
  },
  {
    "id": "de-conversation-04-base17-block-timed-n4",
    "case": "de-conversation-04",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 4,
    "block_size": 5,
    "plain": "Im Skript kann ich neben einen einzelnen Satz eine gewünschte Dauer schreiben. Du machst Witze. Trifft das Modell dann wirklich jede einzelne Millisekunde? Nein. Der Codec arbeitet in Achtzig-Millisekunden-Schritten, und die gewünschte Länge kann verfehlt werden. Danke. Ich wollte gerade mein ganzes Leben auf drei Sekunden zusammenfassen lassen. Wir messen die tatsächliche Dauer und zeigen auch vorzeitige Enden und erreichte Längenlimits. Dann lass diesen Satz ruhig klingen, als würde ein Freund nach den richtigen Worten suchen. Genau darum geht es: eine Absicht vorgeben, das Ergebnis messen und aufmerksam zuhören. Ein angenehm menschlicher Umgang mit einer ziemlich eigenartigen kleinen Maschine. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/0b1b1d6aa428e04f196523800694dfabe813cd115b15f3e244074f0aa99f26ca.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-04-base17-block-timed-u00-c02",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 1036800
      },
      {
        "candidate_id": "de-conversation-04-base17-block-timed-u01-c03",
        "output_offset_sample": 1052160,
        "first_word_output_sample": 1056000,
        "last_word_output_sample": 2154240
      },
      {
        "candidate_id": "de-conversation-04-base17-block-timed-u02-c02",
        "output_offset_sample": 2173440,
        "first_word_output_sample": 2173440,
        "last_word_output_sample": 2664960
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 55.67,
    "samples": 2672160,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-04-base17-block-timed-n4.mp3"
  },
  {
    "id": "de-conversation-04-base17-block-timed-n8",
    "case": "de-conversation-04",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 8,
    "block_size": 5,
    "plain": "Im Skript kann ich neben einen einzelnen Satz eine gewünschte Dauer schreiben. Du machst Witze. Trifft das Modell dann wirklich jede einzelne Millisekunde? Nein. Der Codec arbeitet in Achtzig-Millisekunden-Schritten, und die gewünschte Länge kann verfehlt werden. Danke. Ich wollte gerade mein ganzes Leben auf drei Sekunden zusammenfassen lassen. Wir messen die tatsächliche Dauer und zeigen auch vorzeitige Enden und erreichte Längenlimits. Dann lass diesen Satz ruhig klingen, als würde ein Freund nach den richtigen Worten suchen. Genau darum geht es: eine Absicht vorgeben, das Ergebnis messen und aufmerksam zuhören. Ein angenehm menschlicher Umgang mit einer ziemlich eigenartigen kleinen Maschine. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/e29781a84a58bcdc2d044e73a7da1bd7d67db985a0d35bb05a2de25d364af773.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-04-base17-block-timed-u00-c07",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 1032960
      },
      {
        "candidate_id": "de-conversation-04-base17-block-timed-u01-c03",
        "output_offset_sample": 1048320,
        "first_word_output_sample": 1052160,
        "last_word_output_sample": 2150400
      },
      {
        "candidate_id": "de-conversation-04-base17-block-timed-u02-c02",
        "output_offset_sample": 2169600,
        "first_word_output_sample": 2169600,
        "last_word_output_sample": 2661120
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 55.59,
    "samples": 2668320,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-04-base17-block-timed-n8.mp3"
  },
  {
    "id": "de-conversation-04-base17-segment-free-n1",
    "case": "de-conversation-04",
    "model": "base17",
    "strategy": "segment",
    "timing": "free",
    "best_of": 1,
    "block_size": 1,
    "plain": "Im Skript kann ich neben einen einzelnen Satz eine gewünschte Dauer schreiben. Du machst Witze. Trifft das Modell dann wirklich jede einzelne Millisekunde? Nein. Der Codec arbeitet in Achtzig-Millisekunden-Schritten, und die gewünschte Länge kann verfehlt werden. Danke. Ich wollte gerade mein ganzes Leben auf drei Sekunden zusammenfassen lassen. Wir messen die tatsächliche Dauer und zeigen auch vorzeitige Enden und erreichte Längenlimits. Dann lass diesen Satz ruhig klingen, als würde ein Freund nach den richtigen Worten suchen. Genau darum geht es: eine Absicht vorgeben, das Ergebnis messen und aufmerksam zuhören. Ein angenehm menschlicher Umgang mit einer ziemlich eigenartigen kleinen Maschine. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/d53b24347bba8d352a400104882cb66152db0672f52ca53be357f2aa1046f9e4.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-04-base17-segment-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 360960
      },
      {
        "candidate_id": "de-conversation-04-base17-segment-free-u01-c00",
        "output_offset_sample": 376320,
        "first_word_output_sample": 380160,
        "last_word_output_sample": 672000
      },
      {
        "candidate_id": "de-conversation-04-base17-segment-free-u02-c00",
        "output_offset_sample": 687360,
        "first_word_output_sample": 691200,
        "last_word_output_sample": 998400
      },
      {
        "candidate_id": "de-conversation-04-base17-segment-free-u03-c00",
        "output_offset_sample": 1013760,
        "first_word_output_sample": 1017600,
        "last_word_output_sample": 1290240
      },
      {
        "candidate_id": "de-conversation-04-base17-segment-free-u04-c00",
        "output_offset_sample": 1309440,
        "first_word_output_sample": 1309440,
        "last_word_output_sample": 1754880
      },
      {
        "candidate_id": "de-conversation-04-base17-segment-free-u05-c00",
        "output_offset_sample": 1774080,
        "first_word_output_sample": 1774080,
        "last_word_output_sample": 2081280
      },
      {
        "candidate_id": "de-conversation-04-base17-segment-free-u06-c00",
        "output_offset_sample": 2096640,
        "first_word_output_sample": 2100480,
        "last_word_output_sample": 2434560
      },
      {
        "candidate_id": "de-conversation-04-base17-segment-free-u07-c00",
        "output_offset_sample": 2449920,
        "first_word_output_sample": 2453760,
        "last_word_output_sample": 2741760
      },
      {
        "candidate_id": "de-conversation-04-base17-segment-free-u08-c00",
        "output_offset_sample": 2757120,
        "first_word_output_sample": 2760960,
        "last_word_output_sample": 3087360
      },
      {
        "candidate_id": "de-conversation-04-base17-segment-free-u09-c00",
        "output_offset_sample": 3102720,
        "first_word_output_sample": 3106560,
        "last_word_output_sample": 3575040
      },
      {
        "candidate_id": "de-conversation-04-base17-segment-free-u10-c00",
        "output_offset_sample": 3594240,
        "first_word_output_sample": 3594240,
        "last_word_output_sample": 3893760
      },
      {
        "candidate_id": "de-conversation-04-base17-segment-free-u11-c00",
        "output_offset_sample": 3909120,
        "first_word_output_sample": 3912960,
        "last_word_output_sample": 4304640
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 89.83,
    "samples": 4311840,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-04-base17-segment-free-n1.mp3"
  },
  {
    "id": "de-conversation-04-base17-segment-timed-n1",
    "case": "de-conversation-04",
    "model": "base17",
    "strategy": "segment",
    "timing": "timed",
    "best_of": 1,
    "block_size": 1,
    "plain": "Im Skript kann ich neben einen einzelnen Satz eine gewünschte Dauer schreiben. Du machst Witze. Trifft das Modell dann wirklich jede einzelne Millisekunde? Nein. Der Codec arbeitet in Achtzig-Millisekunden-Schritten, und die gewünschte Länge kann verfehlt werden. Danke. Ich wollte gerade mein ganzes Leben auf drei Sekunden zusammenfassen lassen. Wir messen die tatsächliche Dauer und zeigen auch vorzeitige Enden und erreichte Längenlimits. Dann lass diesen Satz ruhig klingen, als würde ein Freund nach den richtigen Worten suchen. Genau darum geht es: eine Absicht vorgeben, das Ergebnis messen und aufmerksam zuhören. Ein angenehm menschlicher Umgang mit einer ziemlich eigenartigen kleinen Maschine. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/e87adefed8447f70df804b844c3f95eb5a8497a404a78b46bbe8343c5dfa138a.mp3",
    "review_candidates": [
      "de-conversation-04-base17-segment-timed-u07-c00"
    ]
  },
  {
    "id": "de-conversation-04-podcast-block-free-n1",
    "case": "de-conversation-04",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 1,
    "block_size": 5,
    "plain": "Im Skript kann ich neben einen einzelnen Satz eine gewünschte Dauer schreiben. Du machst Witze. Trifft das Modell dann wirklich jede einzelne Millisekunde? Nein. Der Codec arbeitet in Achtzig-Millisekunden-Schritten, und die gewünschte Länge kann verfehlt werden. Danke. Ich wollte gerade mein ganzes Leben auf drei Sekunden zusammenfassen lassen. Wir messen die tatsächliche Dauer und zeigen auch vorzeitige Enden und erreichte Längenlimits. Dann lass diesen Satz ruhig klingen, als würde ein Freund nach den richtigen Worten suchen. Genau darum geht es: eine Absicht vorgeben, das Ergebnis messen und aufmerksam zuhören. Ein angenehm menschlicher Umgang mit einer ziemlich eigenartigen kleinen Maschine. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/c0bafebe3de770fa2e7ba022b8d4478628282ed9bc5c6a4c0294ff4ecbbb4303.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-04-podcast-block-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 1121280
      },
      {
        "candidate_id": "de-conversation-04-podcast-block-free-u01-c00",
        "output_offset_sample": 1140480,
        "first_word_output_sample": 1140480,
        "last_word_output_sample": 2342400
      },
      {
        "candidate_id": "de-conversation-04-podcast-block-free-u02-c00",
        "output_offset_sample": 2361600,
        "first_word_output_sample": 2361600,
        "last_word_output_sample": 2906880
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 60.71,
    "samples": 2914080,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-04-podcast-block-free-n1.mp3"
  },
  {
    "id": "de-conversation-04-podcast-block-free-n4",
    "case": "de-conversation-04",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 4,
    "block_size": 5,
    "plain": "Im Skript kann ich neben einen einzelnen Satz eine gewünschte Dauer schreiben. Du machst Witze. Trifft das Modell dann wirklich jede einzelne Millisekunde? Nein. Der Codec arbeitet in Achtzig-Millisekunden-Schritten, und die gewünschte Länge kann verfehlt werden. Danke. Ich wollte gerade mein ganzes Leben auf drei Sekunden zusammenfassen lassen. Wir messen die tatsächliche Dauer und zeigen auch vorzeitige Enden und erreichte Längenlimits. Dann lass diesen Satz ruhig klingen, als würde ein Freund nach den richtigen Worten suchen. Genau darum geht es: eine Absicht vorgeben, das Ergebnis messen und aufmerksam zuhören. Ein angenehm menschlicher Umgang mit einer ziemlich eigenartigen kleinen Maschine. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/cb78fbe552d75ccefabe11d9bd4f054af4cbe857fab342edd52bbe0b6e1f970d.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-04-podcast-block-free-u00-c01",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 1132800
      },
      {
        "candidate_id": "de-conversation-04-podcast-block-free-u01-c02",
        "output_offset_sample": 1152000,
        "first_word_output_sample": 1152000,
        "last_word_output_sample": 2330880
      },
      {
        "candidate_id": "de-conversation-04-podcast-block-free-u02-c00",
        "output_offset_sample": 2350080,
        "first_word_output_sample": 2350080,
        "last_word_output_sample": 2895360
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 60.47,
    "samples": 2902560,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-04-podcast-block-free-n4.mp3"
  },
  {
    "id": "de-conversation-04-podcast-block-free-n8",
    "case": "de-conversation-04",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 8,
    "block_size": 5,
    "plain": "Im Skript kann ich neben einen einzelnen Satz eine gewünschte Dauer schreiben. Du machst Witze. Trifft das Modell dann wirklich jede einzelne Millisekunde? Nein. Der Codec arbeitet in Achtzig-Millisekunden-Schritten, und die gewünschte Länge kann verfehlt werden. Danke. Ich wollte gerade mein ganzes Leben auf drei Sekunden zusammenfassen lassen. Wir messen die tatsächliche Dauer und zeigen auch vorzeitige Enden und erreichte Längenlimits. Dann lass diesen Satz ruhig klingen, als würde ein Freund nach den richtigen Worten suchen. Genau darum geht es: eine Absicht vorgeben, das Ergebnis messen und aufmerksam zuhören. Ein angenehm menschlicher Umgang mit einer ziemlich eigenartigen kleinen Maschine. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/c84d7de1b100915168464c18ff3cf2d4af0c409d3eeee6377eebfa316d882f61.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-04-podcast-block-free-u00-c01",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 1132800
      },
      {
        "candidate_id": "de-conversation-04-podcast-block-free-u01-c06",
        "output_offset_sample": 1152000,
        "first_word_output_sample": 1152000,
        "last_word_output_sample": 2319360
      },
      {
        "candidate_id": "de-conversation-04-podcast-block-free-u02-c00",
        "output_offset_sample": 2338560,
        "first_word_output_sample": 2338560,
        "last_word_output_sample": 2883840
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 60.23,
    "samples": 2891040,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-04-podcast-block-free-n8.mp3"
  },
  {
    "id": "de-conversation-04-podcast-block-timed-n1",
    "case": "de-conversation-04",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 1,
    "block_size": 5,
    "plain": "Im Skript kann ich neben einen einzelnen Satz eine gewünschte Dauer schreiben. Du machst Witze. Trifft das Modell dann wirklich jede einzelne Millisekunde? Nein. Der Codec arbeitet in Achtzig-Millisekunden-Schritten, und die gewünschte Länge kann verfehlt werden. Danke. Ich wollte gerade mein ganzes Leben auf drei Sekunden zusammenfassen lassen. Wir messen die tatsächliche Dauer und zeigen auch vorzeitige Enden und erreichte Längenlimits. Dann lass diesen Satz ruhig klingen, als würde ein Freund nach den richtigen Worten suchen. Genau darum geht es: eine Absicht vorgeben, das Ergebnis messen und aufmerksam zuhören. Ein angenehm menschlicher Umgang mit einer ziemlich eigenartigen kleinen Maschine. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/efcc8af620848e90fd545278dbe6def993824ccee22eb524dc4de44486d6831f.mp3",
    "review_candidates": [
      "de-conversation-04-podcast-block-timed-u00-c00"
    ]
  },
  {
    "id": "de-conversation-04-podcast-block-timed-n4",
    "case": "de-conversation-04",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 4,
    "block_size": 5,
    "plain": "Im Skript kann ich neben einen einzelnen Satz eine gewünschte Dauer schreiben. Du machst Witze. Trifft das Modell dann wirklich jede einzelne Millisekunde? Nein. Der Codec arbeitet in Achtzig-Millisekunden-Schritten, und die gewünschte Länge kann verfehlt werden. Danke. Ich wollte gerade mein ganzes Leben auf drei Sekunden zusammenfassen lassen. Wir messen die tatsächliche Dauer und zeigen auch vorzeitige Enden und erreichte Längenlimits. Dann lass diesen Satz ruhig klingen, als würde ein Freund nach den richtigen Worten suchen. Genau darum geht es: eine Absicht vorgeben, das Ergebnis messen und aufmerksam zuhören. Ein angenehm menschlicher Umgang mit einer ziemlich eigenartigen kleinen Maschine. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/14c51b7182afd3f8c6163704d04a1719443a80afc82b0fcc6e83fe34f782b496.mp3",
    "review_candidates": [
      "de-conversation-04-podcast-block-timed-u00-c00"
    ]
  },
  {
    "id": "de-conversation-04-podcast-block-timed-n8",
    "case": "de-conversation-04",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 8,
    "block_size": 5,
    "plain": "Im Skript kann ich neben einen einzelnen Satz eine gewünschte Dauer schreiben. Du machst Witze. Trifft das Modell dann wirklich jede einzelne Millisekunde? Nein. Der Codec arbeitet in Achtzig-Millisekunden-Schritten, und die gewünschte Länge kann verfehlt werden. Danke. Ich wollte gerade mein ganzes Leben auf drei Sekunden zusammenfassen lassen. Wir messen die tatsächliche Dauer und zeigen auch vorzeitige Enden und erreichte Längenlimits. Dann lass diesen Satz ruhig klingen, als würde ein Freund nach den richtigen Worten suchen. Genau darum geht es: eine Absicht vorgeben, das Ergebnis messen und aufmerksam zuhören. Ein angenehm menschlicher Umgang mit einer ziemlich eigenartigen kleinen Maschine. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/14c51b7182afd3f8c6163704d04a1719443a80afc82b0fcc6e83fe34f782b496.mp3",
    "review_candidates": [
      "de-conversation-04-podcast-block-timed-u00-c00"
    ]
  },
  {
    "id": "de-conversation-04-podcast-segment-free-n1",
    "case": "de-conversation-04",
    "model": "podcast",
    "strategy": "segment",
    "timing": "free",
    "best_of": 1,
    "block_size": 1,
    "plain": "Im Skript kann ich neben einen einzelnen Satz eine gewünschte Dauer schreiben. Du machst Witze. Trifft das Modell dann wirklich jede einzelne Millisekunde? Nein. Der Codec arbeitet in Achtzig-Millisekunden-Schritten, und die gewünschte Länge kann verfehlt werden. Danke. Ich wollte gerade mein ganzes Leben auf drei Sekunden zusammenfassen lassen. Wir messen die tatsächliche Dauer und zeigen auch vorzeitige Enden und erreichte Längenlimits. Dann lass diesen Satz ruhig klingen, als würde ein Freund nach den richtigen Worten suchen. Genau darum geht es: eine Absicht vorgeben, das Ergebnis messen und aufmerksam zuhören. Ein angenehm menschlicher Umgang mit einer ziemlich eigenartigen kleinen Maschine. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/848dc4b7ab64f6ffe5a3bc1730d27fac2b4d15b69374921236b9d424e7af0549.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-04-podcast-segment-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 253440
      },
      {
        "candidate_id": "de-conversation-04-podcast-segment-free-u01-c00",
        "output_offset_sample": 268800,
        "first_word_output_sample": 272640,
        "last_word_output_sample": 533760
      },
      {
        "candidate_id": "de-conversation-04-podcast-segment-free-u02-c00",
        "output_offset_sample": 552960,
        "first_word_output_sample": 552960,
        "last_word_output_sample": 871680
      },
      {
        "candidate_id": "de-conversation-04-podcast-segment-free-u03-c00",
        "output_offset_sample": 887040,
        "first_word_output_sample": 890880,
        "last_word_output_sample": 1148160
      },
      {
        "candidate_id": "de-conversation-04-podcast-segment-free-u04-c00",
        "output_offset_sample": 1167360,
        "first_word_output_sample": 1167360,
        "last_word_output_sample": 1474560
      },
      {
        "candidate_id": "de-conversation-04-podcast-segment-free-u05-c00",
        "output_offset_sample": 1493760,
        "first_word_output_sample": 1493760,
        "last_word_output_sample": 1762560
      },
      {
        "candidate_id": "de-conversation-04-podcast-segment-free-u06-c00",
        "output_offset_sample": 1777920,
        "first_word_output_sample": 1781760,
        "last_word_output_sample": 2077440
      },
      {
        "candidate_id": "de-conversation-04-podcast-segment-free-u07-c00",
        "output_offset_sample": 2092800,
        "first_word_output_sample": 2096640,
        "last_word_output_sample": 2350080
      },
      {
        "candidate_id": "de-conversation-04-podcast-segment-free-u08-c00",
        "output_offset_sample": 2369280,
        "first_word_output_sample": 2369280,
        "last_word_output_sample": 2680320
      },
      {
        "candidate_id": "de-conversation-04-podcast-segment-free-u09-c00",
        "output_offset_sample": 2695680,
        "first_word_output_sample": 2699520,
        "last_word_output_sample": 2968320
      },
      {
        "candidate_id": "de-conversation-04-podcast-segment-free-u10-c00",
        "output_offset_sample": 2983680,
        "first_word_output_sample": 2987520,
        "last_word_output_sample": 3260160
      },
      {
        "candidate_id": "de-conversation-04-podcast-segment-free-u11-c00",
        "output_offset_sample": 3275520,
        "first_word_output_sample": 3279360,
        "last_word_output_sample": 3575040
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 74.63,
    "samples": 3582240,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-04-podcast-segment-free-n1.mp3"
  },
  {
    "id": "de-conversation-04-podcast-segment-timed-n1",
    "case": "de-conversation-04",
    "model": "podcast",
    "strategy": "segment",
    "timing": "timed",
    "best_of": 1,
    "block_size": 1,
    "plain": "Im Skript kann ich neben einen einzelnen Satz eine gewünschte Dauer schreiben. Du machst Witze. Trifft das Modell dann wirklich jede einzelne Millisekunde? Nein. Der Codec arbeitet in Achtzig-Millisekunden-Schritten, und die gewünschte Länge kann verfehlt werden. Danke. Ich wollte gerade mein ganzes Leben auf drei Sekunden zusammenfassen lassen. Wir messen die tatsächliche Dauer und zeigen auch vorzeitige Enden und erreichte Längenlimits. Dann lass diesen Satz ruhig klingen, als würde ein Freund nach den richtigen Worten suchen. Genau darum geht es: eine Absicht vorgeben, das Ergebnis messen und aufmerksam zuhören. Ein angenehm menschlicher Umgang mit einer ziemlich eigenartigen kleinen Maschine. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/8387476e335cfd7ef47b837a498b21a1e521a7283dc77f69f9baecd4b78afcf4.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-04-podcast-segment-timed-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 207360
      },
      {
        "candidate_id": "de-conversation-04-podcast-segment-timed-u01-c00",
        "output_offset_sample": 222720,
        "first_word_output_sample": 226560,
        "last_word_output_sample": 410880
      },
      {
        "candidate_id": "de-conversation-04-podcast-segment-timed-u02-c00",
        "output_offset_sample": 426240,
        "first_word_output_sample": 430080,
        "last_word_output_sample": 648960
      },
      {
        "candidate_id": "de-conversation-04-podcast-segment-timed-u03-c00",
        "output_offset_sample": 668160,
        "first_word_output_sample": 668160,
        "last_word_output_sample": 875520
      },
      {
        "candidate_id": "de-conversation-04-podcast-segment-timed-u04-c00",
        "output_offset_sample": 894720,
        "first_word_output_sample": 894720,
        "last_word_output_sample": 1117440
      },
      {
        "candidate_id": "de-conversation-04-podcast-segment-timed-u05-c00",
        "output_offset_sample": 1136640,
        "first_word_output_sample": 1136640,
        "last_word_output_sample": 1386240
      },
      {
        "candidate_id": "de-conversation-04-podcast-segment-timed-u06-c00",
        "output_offset_sample": 1405440,
        "first_word_output_sample": 1405440,
        "last_word_output_sample": 1635840
      },
      {
        "candidate_id": "de-conversation-04-podcast-segment-timed-u07-c00",
        "output_offset_sample": 1651200,
        "first_word_output_sample": 1655040,
        "last_word_output_sample": 1824000
      },
      {
        "candidate_id": "de-conversation-04-podcast-segment-timed-u08-c00",
        "output_offset_sample": 1843200,
        "first_word_output_sample": 1843200,
        "last_word_output_sample": 2069760
      },
      {
        "candidate_id": "de-conversation-04-podcast-segment-timed-u09-c00",
        "output_offset_sample": 2088960,
        "first_word_output_sample": 2088960,
        "last_word_output_sample": 2327040
      },
      {
        "candidate_id": "de-conversation-04-podcast-segment-timed-u10-c00",
        "output_offset_sample": 2346240,
        "first_word_output_sample": 2346240,
        "last_word_output_sample": 2618880
      },
      {
        "candidate_id": "de-conversation-04-podcast-segment-timed-u11-c00",
        "output_offset_sample": 2634240,
        "first_word_output_sample": 2638080,
        "last_word_output_sample": 2849280
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 59.51,
    "samples": 2856480,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-04-podcast-segment-timed-n1.mp3"
  },
  {
    "id": "de-conversation-05-base17-block-free-n1",
    "case": "de-conversation-05",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 1,
    "block_size": 4,
    "plain": "Eine Hörerin erlebt erst Englisch, dann Deutsch, Französisch, Spanisch und Niederländisch. Dieselbe Modellfamilie, andere Stimmen und in jeder Sprache ein eigener emotionaler Rhythmus? Das untersucht dieses Hörgrid, mit kurzen Dialogen und längeren Gesprächen. Ich würde ohne Referenz anfangen und zwei Gastgeber erfinden lassen. Gewagt, aber spannend. Danach drei, klar als Versuch gekennzeichnet, weil das Training auf zwei Sprecher ausgerichtet war. Ich wünsche mir Wärme, Überraschung und ein bisschen Neckerei, aber keine vorgelesenen Regieanweisungen. Die vollständigen Skripte und Seeds bleiben sichtbar, damit auch interessante Fehler nachvollziehbar sind. Gut. Ich bin noch skeptisch, aber die nächste Antwort möchte ich jetzt wirklich hören. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht. Stell dir eine kleine Wissenschaftssendung vor, bei der ehrliche Neugier wichtiger ist als große Versprechen. Und die zweite Person darf sagen: Moment, erklär mir das bitte noch einmal, ich habe den Gedanken noch nicht verstanden. Genau. Das Gespräch soll aufeinander reagieren, und nach jedem Stimmungswechsel sollen dieselben zwei Menschen wiederzuerkennen sein. Dann ist es ein fairer Vergleich. Wir lassen die Worte gleich und hören, was die verschiedenen Verfahren daraus machen.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/d4d68d2b92c4d356b5aa7a30c2e31321f38d55aecddda80f1c40c233277a2637.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-05-base17-block-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 948480
      },
      {
        "candidate_id": "de-conversation-05-base17-block-free-u01-c00",
        "output_offset_sample": 963840,
        "first_word_output_sample": 967680,
        "last_word_output_sample": 1923840
      },
      {
        "candidate_id": "de-conversation-05-base17-block-free-u02-c00",
        "output_offset_sample": 1943040,
        "first_word_output_sample": 1943040,
        "last_word_output_sample": 2899200
      },
      {
        "candidate_id": "de-conversation-05-base17-block-free-u03-c00",
        "output_offset_sample": 2918400,
        "first_word_output_sample": 2918400,
        "last_word_output_sample": 3924480
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 81.91,
    "samples": 3931680,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-05-base17-block-free-n1.mp3"
  },
  {
    "id": "de-conversation-05-base17-block-free-n4",
    "case": "de-conversation-05",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 4,
    "block_size": 4,
    "plain": "Eine Hörerin erlebt erst Englisch, dann Deutsch, Französisch, Spanisch und Niederländisch. Dieselbe Modellfamilie, andere Stimmen und in jeder Sprache ein eigener emotionaler Rhythmus? Das untersucht dieses Hörgrid, mit kurzen Dialogen und längeren Gesprächen. Ich würde ohne Referenz anfangen und zwei Gastgeber erfinden lassen. Gewagt, aber spannend. Danach drei, klar als Versuch gekennzeichnet, weil das Training auf zwei Sprecher ausgerichtet war. Ich wünsche mir Wärme, Überraschung und ein bisschen Neckerei, aber keine vorgelesenen Regieanweisungen. Die vollständigen Skripte und Seeds bleiben sichtbar, damit auch interessante Fehler nachvollziehbar sind. Gut. Ich bin noch skeptisch, aber die nächste Antwort möchte ich jetzt wirklich hören. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht. Stell dir eine kleine Wissenschaftssendung vor, bei der ehrliche Neugier wichtiger ist als große Versprechen. Und die zweite Person darf sagen: Moment, erklär mir das bitte noch einmal, ich habe den Gedanken noch nicht verstanden. Genau. Das Gespräch soll aufeinander reagieren, und nach jedem Stimmungswechsel sollen dieselben zwei Menschen wiederzuerkennen sein. Dann ist es ein fairer Vergleich. Wir lassen die Worte gleich und hören, was die verschiedenen Verfahren daraus machen.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/a94ad51472a3af04f3bb1ce810192166389b1bc6ed266efcb31de616fb8c8809.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-05-base17-block-free-u00-c03",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 921600
      },
      {
        "candidate_id": "de-conversation-05-base17-block-free-u01-c00",
        "output_offset_sample": 936960,
        "first_word_output_sample": 940800,
        "last_word_output_sample": 1896960
      },
      {
        "candidate_id": "de-conversation-05-base17-block-free-u02-c00",
        "output_offset_sample": 1916160,
        "first_word_output_sample": 1916160,
        "last_word_output_sample": 2872320
      },
      {
        "candidate_id": "de-conversation-05-base17-block-free-u03-c01",
        "output_offset_sample": 2887680,
        "first_word_output_sample": 2891520,
        "last_word_output_sample": 3951360
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 82.47,
    "samples": 3958560,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-05-base17-block-free-n4.mp3"
  },
  {
    "id": "de-conversation-05-base17-block-free-n8",
    "case": "de-conversation-05",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 8,
    "block_size": 4,
    "plain": "Eine Hörerin erlebt erst Englisch, dann Deutsch, Französisch, Spanisch und Niederländisch. Dieselbe Modellfamilie, andere Stimmen und in jeder Sprache ein eigener emotionaler Rhythmus? Das untersucht dieses Hörgrid, mit kurzen Dialogen und längeren Gesprächen. Ich würde ohne Referenz anfangen und zwei Gastgeber erfinden lassen. Gewagt, aber spannend. Danach drei, klar als Versuch gekennzeichnet, weil das Training auf zwei Sprecher ausgerichtet war. Ich wünsche mir Wärme, Überraschung und ein bisschen Neckerei, aber keine vorgelesenen Regieanweisungen. Die vollständigen Skripte und Seeds bleiben sichtbar, damit auch interessante Fehler nachvollziehbar sind. Gut. Ich bin noch skeptisch, aber die nächste Antwort möchte ich jetzt wirklich hören. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht. Stell dir eine kleine Wissenschaftssendung vor, bei der ehrliche Neugier wichtiger ist als große Versprechen. Und die zweite Person darf sagen: Moment, erklär mir das bitte noch einmal, ich habe den Gedanken noch nicht verstanden. Genau. Das Gespräch soll aufeinander reagieren, und nach jedem Stimmungswechsel sollen dieselben zwei Menschen wiederzuerkennen sein. Dann ist es ein fairer Vergleich. Wir lassen die Worte gleich und hören, was die verschiedenen Verfahren daraus machen.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/024fa2977a8e8cf9066ecef892fabd7338d5a9ab088abc6b4a51e834ade757f4.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-05-base17-block-free-u00-c03",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 921600
      },
      {
        "candidate_id": "de-conversation-05-base17-block-free-u01-c00",
        "output_offset_sample": 936960,
        "first_word_output_sample": 940800,
        "last_word_output_sample": 1896960
      },
      {
        "candidate_id": "de-conversation-05-base17-block-free-u02-c00",
        "output_offset_sample": 1916160,
        "first_word_output_sample": 1916160,
        "last_word_output_sample": 2872320
      },
      {
        "candidate_id": "de-conversation-05-base17-block-free-u03-c04",
        "output_offset_sample": 2887680,
        "first_word_output_sample": 2891520,
        "last_word_output_sample": 3912960
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 81.67,
    "samples": 3920160,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-05-base17-block-free-n8.mp3"
  },
  {
    "id": "de-conversation-05-base17-block-timed-n1",
    "case": "de-conversation-05",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 1,
    "block_size": 4,
    "plain": "Eine Hörerin erlebt erst Englisch, dann Deutsch, Französisch, Spanisch und Niederländisch. Dieselbe Modellfamilie, andere Stimmen und in jeder Sprache ein eigener emotionaler Rhythmus? Das untersucht dieses Hörgrid, mit kurzen Dialogen und längeren Gesprächen. Ich würde ohne Referenz anfangen und zwei Gastgeber erfinden lassen. Gewagt, aber spannend. Danach drei, klar als Versuch gekennzeichnet, weil das Training auf zwei Sprecher ausgerichtet war. Ich wünsche mir Wärme, Überraschung und ein bisschen Neckerei, aber keine vorgelesenen Regieanweisungen. Die vollständigen Skripte und Seeds bleiben sichtbar, damit auch interessante Fehler nachvollziehbar sind. Gut. Ich bin noch skeptisch, aber die nächste Antwort möchte ich jetzt wirklich hören. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht. Stell dir eine kleine Wissenschaftssendung vor, bei der ehrliche Neugier wichtiger ist als große Versprechen. Und die zweite Person darf sagen: Moment, erklär mir das bitte noch einmal, ich habe den Gedanken noch nicht verstanden. Genau. Das Gespräch soll aufeinander reagieren, und nach jedem Stimmungswechsel sollen dieselben zwei Menschen wiederzuerkennen sein. Dann ist es ein fairer Vergleich. Wir lassen die Worte gleich und hören, was die verschiedenen Verfahren daraus machen.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/0aec1bc32db1b546125aec04d5be810d48eb951a0870dc7a90f0f2fa7d396c13.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-05-base17-block-timed-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 779520
      },
      {
        "candidate_id": "de-conversation-05-base17-block-timed-u01-c00",
        "output_offset_sample": 794880,
        "first_word_output_sample": 798720,
        "last_word_output_sample": 1712640
      },
      {
        "candidate_id": "de-conversation-05-base17-block-timed-u02-c00",
        "output_offset_sample": 1728000,
        "first_word_output_sample": 1731840,
        "last_word_output_sample": 2676480
      },
      {
        "candidate_id": "de-conversation-05-base17-block-timed-u03-c00",
        "output_offset_sample": 2691840,
        "first_word_output_sample": 2695680,
        "last_word_output_sample": 3874560
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 80.87,
    "samples": 3881760,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-05-base17-block-timed-n1.mp3"
  },
  {
    "id": "de-conversation-05-base17-block-timed-n4",
    "case": "de-conversation-05",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 4,
    "block_size": 4,
    "plain": "Eine Hörerin erlebt erst Englisch, dann Deutsch, Französisch, Spanisch und Niederländisch. Dieselbe Modellfamilie, andere Stimmen und in jeder Sprache ein eigener emotionaler Rhythmus? Das untersucht dieses Hörgrid, mit kurzen Dialogen und längeren Gesprächen. Ich würde ohne Referenz anfangen und zwei Gastgeber erfinden lassen. Gewagt, aber spannend. Danach drei, klar als Versuch gekennzeichnet, weil das Training auf zwei Sprecher ausgerichtet war. Ich wünsche mir Wärme, Überraschung und ein bisschen Neckerei, aber keine vorgelesenen Regieanweisungen. Die vollständigen Skripte und Seeds bleiben sichtbar, damit auch interessante Fehler nachvollziehbar sind. Gut. Ich bin noch skeptisch, aber die nächste Antwort möchte ich jetzt wirklich hören. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht. Stell dir eine kleine Wissenschaftssendung vor, bei der ehrliche Neugier wichtiger ist als große Versprechen. Und die zweite Person darf sagen: Moment, erklär mir das bitte noch einmal, ich habe den Gedanken noch nicht verstanden. Genau. Das Gespräch soll aufeinander reagieren, und nach jedem Stimmungswechsel sollen dieselben zwei Menschen wiederzuerkennen sein. Dann ist es ein fairer Vergleich. Wir lassen die Worte gleich und hören, was die verschiedenen Verfahren daraus machen.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/f5702a69f044fb725ea1d0ce05ad4b01639f85e0ec2eaba7635814d5385a1ab6.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-05-base17-block-timed-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 779520
      },
      {
        "candidate_id": "de-conversation-05-base17-block-timed-u01-c01",
        "output_offset_sample": 794880,
        "first_word_output_sample": 798720,
        "last_word_output_sample": 1708800
      },
      {
        "candidate_id": "de-conversation-05-base17-block-timed-u02-c03",
        "output_offset_sample": 1724160,
        "first_word_output_sample": 1728000,
        "last_word_output_sample": 2672640
      },
      {
        "candidate_id": "de-conversation-05-base17-block-timed-u03-c00",
        "output_offset_sample": 2688000,
        "first_word_output_sample": 2691840,
        "last_word_output_sample": 3870720
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 80.79,
    "samples": 3877920,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-05-base17-block-timed-n4.mp3"
  },
  {
    "id": "de-conversation-05-base17-block-timed-n8",
    "case": "de-conversation-05",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 8,
    "block_size": 4,
    "plain": "Eine Hörerin erlebt erst Englisch, dann Deutsch, Französisch, Spanisch und Niederländisch. Dieselbe Modellfamilie, andere Stimmen und in jeder Sprache ein eigener emotionaler Rhythmus? Das untersucht dieses Hörgrid, mit kurzen Dialogen und längeren Gesprächen. Ich würde ohne Referenz anfangen und zwei Gastgeber erfinden lassen. Gewagt, aber spannend. Danach drei, klar als Versuch gekennzeichnet, weil das Training auf zwei Sprecher ausgerichtet war. Ich wünsche mir Wärme, Überraschung und ein bisschen Neckerei, aber keine vorgelesenen Regieanweisungen. Die vollständigen Skripte und Seeds bleiben sichtbar, damit auch interessante Fehler nachvollziehbar sind. Gut. Ich bin noch skeptisch, aber die nächste Antwort möchte ich jetzt wirklich hören. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht. Stell dir eine kleine Wissenschaftssendung vor, bei der ehrliche Neugier wichtiger ist als große Versprechen. Und die zweite Person darf sagen: Moment, erklär mir das bitte noch einmal, ich habe den Gedanken noch nicht verstanden. Genau. Das Gespräch soll aufeinander reagieren, und nach jedem Stimmungswechsel sollen dieselben zwei Menschen wiederzuerkennen sein. Dann ist es ein fairer Vergleich. Wir lassen die Worte gleich und hören, was die verschiedenen Verfahren daraus machen.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/1848c80b0548e2847764708d3abc2483147516743cc244a4ad2bcb4ccecf3ee2.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-05-base17-block-timed-u00-c07",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 783360
      },
      {
        "candidate_id": "de-conversation-05-base17-block-timed-u01-c04",
        "output_offset_sample": 798720,
        "first_word_output_sample": 802560,
        "last_word_output_sample": 1712640
      },
      {
        "candidate_id": "de-conversation-05-base17-block-timed-u02-c03",
        "output_offset_sample": 1728000,
        "first_word_output_sample": 1731840,
        "last_word_output_sample": 2676480
      },
      {
        "candidate_id": "de-conversation-05-base17-block-timed-u03-c04",
        "output_offset_sample": 2691840,
        "first_word_output_sample": 2695680,
        "last_word_output_sample": 3874560
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 80.87,
    "samples": 3881760,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-05-base17-block-timed-n8.mp3"
  },
  {
    "id": "de-conversation-05-base17-segment-free-n1",
    "case": "de-conversation-05",
    "model": "base17",
    "strategy": "segment",
    "timing": "free",
    "best_of": 1,
    "block_size": 1,
    "plain": "Eine Hörerin erlebt erst Englisch, dann Deutsch, Französisch, Spanisch und Niederländisch. Dieselbe Modellfamilie, andere Stimmen und in jeder Sprache ein eigener emotionaler Rhythmus? Das untersucht dieses Hörgrid, mit kurzen Dialogen und längeren Gesprächen. Ich würde ohne Referenz anfangen und zwei Gastgeber erfinden lassen. Gewagt, aber spannend. Danach drei, klar als Versuch gekennzeichnet, weil das Training auf zwei Sprecher ausgerichtet war. Ich wünsche mir Wärme, Überraschung und ein bisschen Neckerei, aber keine vorgelesenen Regieanweisungen. Die vollständigen Skripte und Seeds bleiben sichtbar, damit auch interessante Fehler nachvollziehbar sind. Gut. Ich bin noch skeptisch, aber die nächste Antwort möchte ich jetzt wirklich hören. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht. Stell dir eine kleine Wissenschaftssendung vor, bei der ehrliche Neugier wichtiger ist als große Versprechen. Und die zweite Person darf sagen: Moment, erklär mir das bitte noch einmal, ich habe den Gedanken noch nicht verstanden. Genau. Das Gespräch soll aufeinander reagieren, und nach jedem Stimmungswechsel sollen dieselben zwei Menschen wiederzuerkennen sein. Dann ist es ein fairer Vergleich. Wir lassen die Worte gleich und hören, was die verschiedenen Verfahren daraus machen.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": null,
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-05-base17-segment-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 345600
      },
      {
        "candidate_id": "de-conversation-05-base17-segment-free-u01-c00",
        "output_offset_sample": 360960,
        "first_word_output_sample": 364800,
        "last_word_output_sample": 764160
      },
      {
        "candidate_id": "de-conversation-05-base17-segment-free-u02-c00",
        "output_offset_sample": 783360,
        "first_word_output_sample": 783360,
        "last_word_output_sample": 1002240
      },
      {
        "candidate_id": "de-conversation-05-base17-segment-free-u03-c00",
        "output_offset_sample": 1017600,
        "first_word_output_sample": 1021440,
        "last_word_output_sample": 1420800
      },
      {
        "candidate_id": "de-conversation-05-base17-segment-free-u04-c00",
        "output_offset_sample": 1436160,
        "first_word_output_sample": 1440000,
        "last_word_output_sample": 1770240
      },
      {
        "candidate_id": "de-conversation-05-base17-segment-free-u05-c00",
        "output_offset_sample": 1789440,
        "first_word_output_sample": 1789440,
        "last_word_output_sample": 2161920
      },
      {
        "candidate_id": "de-conversation-05-base17-segment-free-u06-c00",
        "output_offset_sample": 2181120,
        "first_word_output_sample": 2181120,
        "last_word_output_sample": 2565120
      },
      {
        "candidate_id": "de-conversation-05-base17-segment-free-u07-c00",
        "output_offset_sample": 2584320,
        "first_word_output_sample": 2584320,
        "last_word_output_sample": 2845440
      },
      {
        "candidate_id": "de-conversation-05-base17-segment-free-u08-c00",
        "output_offset_sample": 2864640,
        "first_word_output_sample": 2864640,
        "last_word_output_sample": 3187200
      },
      {
        "candidate_id": "de-conversation-05-base17-segment-free-u09-c00",
        "output_offset_sample": 3202560,
        "first_word_output_sample": 3206400,
        "last_word_output_sample": 3632640
      },
      {
        "candidate_id": "de-conversation-05-base17-segment-free-u10-c00",
        "output_offset_sample": 3651840,
        "first_word_output_sample": 3651840,
        "last_word_output_sample": 3951360
      },
      {
        "candidate_id": "de-conversation-05-base17-segment-free-u11-c00",
        "output_offset_sample": 3966720,
        "first_word_output_sample": 3970560,
        "last_word_output_sample": 4273920
      },
      {
        "candidate_id": "de-conversation-05-base17-segment-free-u12-c00",
        "output_offset_sample": 4289280,
        "first_word_output_sample": 4293120,
        "last_word_output_sample": 4638720
      },
      {
        "candidate_id": "de-conversation-05-base17-segment-free-u13-c00",
        "output_offset_sample": 4654080,
        "first_word_output_sample": 4657920,
        "last_word_output_sample": 5034240
      },
      {
        "candidate_id": "de-conversation-05-base17-segment-free-u14-c00",
        "output_offset_sample": 5049600,
        "first_word_output_sample": 5053440,
        "last_word_output_sample": 5406720
      },
      {
        "candidate_id": "de-conversation-05-base17-segment-free-u15-c00",
        "output_offset_sample": 5425920,
        "first_word_output_sample": 5425920,
        "last_word_output_sample": 5736960
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 119.67,
    "samples": 5744160,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-05-base17-segment-free-n1.mp3"
  },
  {
    "id": "de-conversation-05-base17-segment-timed-n1",
    "case": "de-conversation-05",
    "model": "base17",
    "strategy": "segment",
    "timing": "timed",
    "best_of": 1,
    "block_size": 1,
    "plain": "Eine Hörerin erlebt erst Englisch, dann Deutsch, Französisch, Spanisch und Niederländisch. Dieselbe Modellfamilie, andere Stimmen und in jeder Sprache ein eigener emotionaler Rhythmus? Das untersucht dieses Hörgrid, mit kurzen Dialogen und längeren Gesprächen. Ich würde ohne Referenz anfangen und zwei Gastgeber erfinden lassen. Gewagt, aber spannend. Danach drei, klar als Versuch gekennzeichnet, weil das Training auf zwei Sprecher ausgerichtet war. Ich wünsche mir Wärme, Überraschung und ein bisschen Neckerei, aber keine vorgelesenen Regieanweisungen. Die vollständigen Skripte und Seeds bleiben sichtbar, damit auch interessante Fehler nachvollziehbar sind. Gut. Ich bin noch skeptisch, aber die nächste Antwort möchte ich jetzt wirklich hören. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht. Stell dir eine kleine Wissenschaftssendung vor, bei der ehrliche Neugier wichtiger ist als große Versprechen. Und die zweite Person darf sagen: Moment, erklär mir das bitte noch einmal, ich habe den Gedanken noch nicht verstanden. Genau. Das Gespräch soll aufeinander reagieren, und nach jedem Stimmungswechsel sollen dieselben zwei Menschen wiederzuerkennen sein. Dann ist es ein fairer Vergleich. Wir lassen die Worte gleich und hören, was die verschiedenen Verfahren daraus machen.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/27f6a493a15432f8cfd597733c77fc3ea679bd2943bcfe67d20571781ba22718.mp3",
    "review_candidates": [
      "de-conversation-05-base17-segment-timed-u00-c00",
      "de-conversation-05-base17-segment-timed-u01-c00",
      "de-conversation-05-base17-segment-timed-u05-c00"
    ]
  },
  {
    "id": "de-conversation-05-podcast-block-free-n1",
    "case": "de-conversation-05",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 1,
    "block_size": 4,
    "plain": "Eine Hörerin erlebt erst Englisch, dann Deutsch, Französisch, Spanisch und Niederländisch. Dieselbe Modellfamilie, andere Stimmen und in jeder Sprache ein eigener emotionaler Rhythmus? Das untersucht dieses Hörgrid, mit kurzen Dialogen und längeren Gesprächen. Ich würde ohne Referenz anfangen und zwei Gastgeber erfinden lassen. Gewagt, aber spannend. Danach drei, klar als Versuch gekennzeichnet, weil das Training auf zwei Sprecher ausgerichtet war. Ich wünsche mir Wärme, Überraschung und ein bisschen Neckerei, aber keine vorgelesenen Regieanweisungen. Die vollständigen Skripte und Seeds bleiben sichtbar, damit auch interessante Fehler nachvollziehbar sind. Gut. Ich bin noch skeptisch, aber die nächste Antwort möchte ich jetzt wirklich hören. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht. Stell dir eine kleine Wissenschaftssendung vor, bei der ehrliche Neugier wichtiger ist als große Versprechen. Und die zweite Person darf sagen: Moment, erklär mir das bitte noch einmal, ich habe den Gedanken noch nicht verstanden. Genau. Das Gespräch soll aufeinander reagieren, und nach jedem Stimmungswechsel sollen dieselben zwei Menschen wiederzuerkennen sein. Dann ist es ein fairer Vergleich. Wir lassen die Worte gleich und hören, was die verschiedenen Verfahren daraus machen.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": null,
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-05-podcast-block-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 990720
      },
      {
        "candidate_id": "de-conversation-05-podcast-block-free-u01-c00",
        "output_offset_sample": 1006080,
        "first_word_output_sample": 1009920,
        "last_word_output_sample": 2050560
      },
      {
        "candidate_id": "de-conversation-05-podcast-block-free-u02-c00",
        "output_offset_sample": 2069760,
        "first_word_output_sample": 2069760,
        "last_word_output_sample": 3091200
      },
      {
        "candidate_id": "de-conversation-05-podcast-block-free-u03-c00",
        "output_offset_sample": 3110400,
        "first_word_output_sample": 3110400,
        "last_word_output_sample": 4239360
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 88.47,
    "samples": 4246560,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-05-podcast-block-free-n1.mp3"
  },
  {
    "id": "de-conversation-05-podcast-block-free-n4",
    "case": "de-conversation-05",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 4,
    "block_size": 4,
    "plain": "Eine Hörerin erlebt erst Englisch, dann Deutsch, Französisch, Spanisch und Niederländisch. Dieselbe Modellfamilie, andere Stimmen und in jeder Sprache ein eigener emotionaler Rhythmus? Das untersucht dieses Hörgrid, mit kurzen Dialogen und längeren Gesprächen. Ich würde ohne Referenz anfangen und zwei Gastgeber erfinden lassen. Gewagt, aber spannend. Danach drei, klar als Versuch gekennzeichnet, weil das Training auf zwei Sprecher ausgerichtet war. Ich wünsche mir Wärme, Überraschung und ein bisschen Neckerei, aber keine vorgelesenen Regieanweisungen. Die vollständigen Skripte und Seeds bleiben sichtbar, damit auch interessante Fehler nachvollziehbar sind. Gut. Ich bin noch skeptisch, aber die nächste Antwort möchte ich jetzt wirklich hören. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht. Stell dir eine kleine Wissenschaftssendung vor, bei der ehrliche Neugier wichtiger ist als große Versprechen. Und die zweite Person darf sagen: Moment, erklär mir das bitte noch einmal, ich habe den Gedanken noch nicht verstanden. Genau. Das Gespräch soll aufeinander reagieren, und nach jedem Stimmungswechsel sollen dieselben zwei Menschen wiederzuerkennen sein. Dann ist es ein fairer Vergleich. Wir lassen die Worte gleich und hören, was die verschiedenen Verfahren daraus machen.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": null,
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-05-podcast-block-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 990720
      },
      {
        "candidate_id": "de-conversation-05-podcast-block-free-u01-c00",
        "output_offset_sample": 1006080,
        "first_word_output_sample": 1009920,
        "last_word_output_sample": 2050560
      },
      {
        "candidate_id": "de-conversation-05-podcast-block-free-u02-c02",
        "output_offset_sample": 2069760,
        "first_word_output_sample": 2069760,
        "last_word_output_sample": 3083520
      },
      {
        "candidate_id": "de-conversation-05-podcast-block-free-u03-c01",
        "output_offset_sample": 3102720,
        "first_word_output_sample": 3102720,
        "last_word_output_sample": 4235520
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 88.39,
    "samples": 4242720,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-05-podcast-block-free-n4.mp3"
  },
  {
    "id": "de-conversation-05-podcast-block-free-n8",
    "case": "de-conversation-05",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 8,
    "block_size": 4,
    "plain": "Eine Hörerin erlebt erst Englisch, dann Deutsch, Französisch, Spanisch und Niederländisch. Dieselbe Modellfamilie, andere Stimmen und in jeder Sprache ein eigener emotionaler Rhythmus? Das untersucht dieses Hörgrid, mit kurzen Dialogen und längeren Gesprächen. Ich würde ohne Referenz anfangen und zwei Gastgeber erfinden lassen. Gewagt, aber spannend. Danach drei, klar als Versuch gekennzeichnet, weil das Training auf zwei Sprecher ausgerichtet war. Ich wünsche mir Wärme, Überraschung und ein bisschen Neckerei, aber keine vorgelesenen Regieanweisungen. Die vollständigen Skripte und Seeds bleiben sichtbar, damit auch interessante Fehler nachvollziehbar sind. Gut. Ich bin noch skeptisch, aber die nächste Antwort möchte ich jetzt wirklich hören. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht. Stell dir eine kleine Wissenschaftssendung vor, bei der ehrliche Neugier wichtiger ist als große Versprechen. Und die zweite Person darf sagen: Moment, erklär mir das bitte noch einmal, ich habe den Gedanken noch nicht verstanden. Genau. Das Gespräch soll aufeinander reagieren, und nach jedem Stimmungswechsel sollen dieselben zwei Menschen wiederzuerkennen sein. Dann ist es ein fairer Vergleich. Wir lassen die Worte gleich und hören, was die verschiedenen Verfahren daraus machen.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": null,
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-05-podcast-block-free-u00-c06",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 990720
      },
      {
        "candidate_id": "de-conversation-05-podcast-block-free-u01-c00",
        "output_offset_sample": 1006080,
        "first_word_output_sample": 1009920,
        "last_word_output_sample": 2050560
      },
      {
        "candidate_id": "de-conversation-05-podcast-block-free-u02-c07",
        "output_offset_sample": 2069760,
        "first_word_output_sample": 2069760,
        "last_word_output_sample": 3106560
      },
      {
        "candidate_id": "de-conversation-05-podcast-block-free-u03-c07",
        "output_offset_sample": 3125760,
        "first_word_output_sample": 3125760,
        "last_word_output_sample": 4266240
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 89.03,
    "samples": 4273440,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-05-podcast-block-free-n8.mp3"
  },
  {
    "id": "de-conversation-05-podcast-block-timed-n1",
    "case": "de-conversation-05",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 1,
    "block_size": 4,
    "plain": "Eine Hörerin erlebt erst Englisch, dann Deutsch, Französisch, Spanisch und Niederländisch. Dieselbe Modellfamilie, andere Stimmen und in jeder Sprache ein eigener emotionaler Rhythmus? Das untersucht dieses Hörgrid, mit kurzen Dialogen und längeren Gesprächen. Ich würde ohne Referenz anfangen und zwei Gastgeber erfinden lassen. Gewagt, aber spannend. Danach drei, klar als Versuch gekennzeichnet, weil das Training auf zwei Sprecher ausgerichtet war. Ich wünsche mir Wärme, Überraschung und ein bisschen Neckerei, aber keine vorgelesenen Regieanweisungen. Die vollständigen Skripte und Seeds bleiben sichtbar, damit auch interessante Fehler nachvollziehbar sind. Gut. Ich bin noch skeptisch, aber die nächste Antwort möchte ich jetzt wirklich hören. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht. Stell dir eine kleine Wissenschaftssendung vor, bei der ehrliche Neugier wichtiger ist als große Versprechen. Und die zweite Person darf sagen: Moment, erklär mir das bitte noch einmal, ich habe den Gedanken noch nicht verstanden. Genau. Das Gespräch soll aufeinander reagieren, und nach jedem Stimmungswechsel sollen dieselben zwei Menschen wiederzuerkennen sein. Dann ist es ein fairer Vergleich. Wir lassen die Worte gleich und hören, was die verschiedenen Verfahren daraus machen.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/1cda076b2ce91fbfb40d4f2d14eec79223a8f20c0d9ac599919222dc6f605203.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-05-podcast-block-timed-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 779520
      },
      {
        "candidate_id": "de-conversation-05-podcast-block-timed-u01-c00",
        "output_offset_sample": 794880,
        "first_word_output_sample": 798720,
        "last_word_output_sample": 1704960
      },
      {
        "candidate_id": "de-conversation-05-podcast-block-timed-u02-c00",
        "output_offset_sample": 1724160,
        "first_word_output_sample": 1724160,
        "last_word_output_sample": 2680320
      },
      {
        "candidate_id": "de-conversation-05-podcast-block-timed-u03-c00",
        "output_offset_sample": 2699520,
        "first_word_output_sample": 2699520,
        "last_word_output_sample": 3889920
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 81.19,
    "samples": 3897120,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-05-podcast-block-timed-n1.mp3"
  },
  {
    "id": "de-conversation-05-podcast-block-timed-n4",
    "case": "de-conversation-05",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 4,
    "block_size": 4,
    "plain": "Eine Hörerin erlebt erst Englisch, dann Deutsch, Französisch, Spanisch und Niederländisch. Dieselbe Modellfamilie, andere Stimmen und in jeder Sprache ein eigener emotionaler Rhythmus? Das untersucht dieses Hörgrid, mit kurzen Dialogen und längeren Gesprächen. Ich würde ohne Referenz anfangen und zwei Gastgeber erfinden lassen. Gewagt, aber spannend. Danach drei, klar als Versuch gekennzeichnet, weil das Training auf zwei Sprecher ausgerichtet war. Ich wünsche mir Wärme, Überraschung und ein bisschen Neckerei, aber keine vorgelesenen Regieanweisungen. Die vollständigen Skripte und Seeds bleiben sichtbar, damit auch interessante Fehler nachvollziehbar sind. Gut. Ich bin noch skeptisch, aber die nächste Antwort möchte ich jetzt wirklich hören. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht. Stell dir eine kleine Wissenschaftssendung vor, bei der ehrliche Neugier wichtiger ist als große Versprechen. Und die zweite Person darf sagen: Moment, erklär mir das bitte noch einmal, ich habe den Gedanken noch nicht verstanden. Genau. Das Gespräch soll aufeinander reagieren, und nach jedem Stimmungswechsel sollen dieselben zwei Menschen wiederzuerkennen sein. Dann ist es ein fairer Vergleich. Wir lassen die Worte gleich und hören, was die verschiedenen Verfahren daraus machen.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/34e0b85466bcaa37e035efcb324d25e341d684b75e8f2da4fda42cd9ef451960.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-05-podcast-block-timed-u00-c02",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 783360
      },
      {
        "candidate_id": "de-conversation-05-podcast-block-timed-u01-c01",
        "output_offset_sample": 798720,
        "first_word_output_sample": 802560,
        "last_word_output_sample": 1716480
      },
      {
        "candidate_id": "de-conversation-05-podcast-block-timed-u02-c02",
        "output_offset_sample": 1735680,
        "first_word_output_sample": 1735680,
        "last_word_output_sample": 2691840
      },
      {
        "candidate_id": "de-conversation-05-podcast-block-timed-u03-c00",
        "output_offset_sample": 2711040,
        "first_word_output_sample": 2711040,
        "last_word_output_sample": 3901440
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 81.43,
    "samples": 3908640,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-05-podcast-block-timed-n4.mp3"
  },
  {
    "id": "de-conversation-05-podcast-block-timed-n8",
    "case": "de-conversation-05",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 8,
    "block_size": 4,
    "plain": "Eine Hörerin erlebt erst Englisch, dann Deutsch, Französisch, Spanisch und Niederländisch. Dieselbe Modellfamilie, andere Stimmen und in jeder Sprache ein eigener emotionaler Rhythmus? Das untersucht dieses Hörgrid, mit kurzen Dialogen und längeren Gesprächen. Ich würde ohne Referenz anfangen und zwei Gastgeber erfinden lassen. Gewagt, aber spannend. Danach drei, klar als Versuch gekennzeichnet, weil das Training auf zwei Sprecher ausgerichtet war. Ich wünsche mir Wärme, Überraschung und ein bisschen Neckerei, aber keine vorgelesenen Regieanweisungen. Die vollständigen Skripte und Seeds bleiben sichtbar, damit auch interessante Fehler nachvollziehbar sind. Gut. Ich bin noch skeptisch, aber die nächste Antwort möchte ich jetzt wirklich hören. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht. Stell dir eine kleine Wissenschaftssendung vor, bei der ehrliche Neugier wichtiger ist als große Versprechen. Und die zweite Person darf sagen: Moment, erklär mir das bitte noch einmal, ich habe den Gedanken noch nicht verstanden. Genau. Das Gespräch soll aufeinander reagieren, und nach jedem Stimmungswechsel sollen dieselben zwei Menschen wiederzuerkennen sein. Dann ist es ein fairer Vergleich. Wir lassen die Worte gleich und hören, was die verschiedenen Verfahren daraus machen.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/34e0b85466bcaa37e035efcb324d25e341d684b75e8f2da4fda42cd9ef451960.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-05-podcast-block-timed-u00-c02",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 783360
      },
      {
        "candidate_id": "de-conversation-05-podcast-block-timed-u01-c01",
        "output_offset_sample": 798720,
        "first_word_output_sample": 802560,
        "last_word_output_sample": 1716480
      },
      {
        "candidate_id": "de-conversation-05-podcast-block-timed-u02-c02",
        "output_offset_sample": 1735680,
        "first_word_output_sample": 1735680,
        "last_word_output_sample": 2691840
      },
      {
        "candidate_id": "de-conversation-05-podcast-block-timed-u03-c00",
        "output_offset_sample": 2711040,
        "first_word_output_sample": 2711040,
        "last_word_output_sample": 3901440
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 81.43,
    "samples": 3908640,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-05-podcast-block-timed-n8.mp3"
  },
  {
    "id": "de-conversation-05-podcast-segment-free-n1",
    "case": "de-conversation-05",
    "model": "podcast",
    "strategy": "segment",
    "timing": "free",
    "best_of": 1,
    "block_size": 1,
    "plain": "Eine Hörerin erlebt erst Englisch, dann Deutsch, Französisch, Spanisch und Niederländisch. Dieselbe Modellfamilie, andere Stimmen und in jeder Sprache ein eigener emotionaler Rhythmus? Das untersucht dieses Hörgrid, mit kurzen Dialogen und längeren Gesprächen. Ich würde ohne Referenz anfangen und zwei Gastgeber erfinden lassen. Gewagt, aber spannend. Danach drei, klar als Versuch gekennzeichnet, weil das Training auf zwei Sprecher ausgerichtet war. Ich wünsche mir Wärme, Überraschung und ein bisschen Neckerei, aber keine vorgelesenen Regieanweisungen. Die vollständigen Skripte und Seeds bleiben sichtbar, damit auch interessante Fehler nachvollziehbar sind. Gut. Ich bin noch skeptisch, aber die nächste Antwort möchte ich jetzt wirklich hören. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht. Stell dir eine kleine Wissenschaftssendung vor, bei der ehrliche Neugier wichtiger ist als große Versprechen. Und die zweite Person darf sagen: Moment, erklär mir das bitte noch einmal, ich habe den Gedanken noch nicht verstanden. Genau. Das Gespräch soll aufeinander reagieren, und nach jedem Stimmungswechsel sollen dieselben zwei Menschen wiederzuerkennen sein. Dann ist es ein fairer Vergleich. Wir lassen die Worte gleich und hören, was die verschiedenen Verfahren daraus machen.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/653c9de2c6d977d7c2fb18c54f73129dd91548bfdc520e2c0d9c780c226e745b.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "de-conversation-05-podcast-segment-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 322560
      },
      {
        "candidate_id": "de-conversation-05-podcast-segment-free-u01-c00",
        "output_offset_sample": 337920,
        "first_word_output_sample": 341760,
        "last_word_output_sample": 652800
      },
      {
        "candidate_id": "de-conversation-05-podcast-segment-free-u02-c00",
        "output_offset_sample": 668160,
        "first_word_output_sample": 672000,
        "last_word_output_sample": 936960
      },
      {
        "candidate_id": "de-conversation-05-podcast-segment-free-u03-c00",
        "output_offset_sample": 952320,
        "first_word_output_sample": 956160,
        "last_word_output_sample": 1271040
      },
      {
        "candidate_id": "de-conversation-05-podcast-segment-free-u04-c00",
        "output_offset_sample": 1286400,
        "first_word_output_sample": 1290240,
        "last_word_output_sample": 1666560
      },
      {
        "candidate_id": "de-conversation-05-podcast-segment-free-u05-c00",
        "output_offset_sample": 1685760,
        "first_word_output_sample": 1685760,
        "last_word_output_sample": 2062080
      },
      {
        "candidate_id": "de-conversation-05-podcast-segment-free-u06-c00",
        "output_offset_sample": 2077440,
        "first_word_output_sample": 2081280,
        "last_word_output_sample": 2376960
      },
      {
        "candidate_id": "de-conversation-05-podcast-segment-free-u07-c00",
        "output_offset_sample": 2396160,
        "first_word_output_sample": 2396160,
        "last_word_output_sample": 2668800
      },
      {
        "candidate_id": "de-conversation-05-podcast-segment-free-u08-c00",
        "output_offset_sample": 2684160,
        "first_word_output_sample": 2688000,
        "last_word_output_sample": 2991360
      },
      {
        "candidate_id": "de-conversation-05-podcast-segment-free-u09-c00",
        "output_offset_sample": 3006720,
        "first_word_output_sample": 3010560,
        "last_word_output_sample": 3352320
      },
      {
        "candidate_id": "de-conversation-05-podcast-segment-free-u10-c00",
        "output_offset_sample": 3371520,
        "first_word_output_sample": 3371520,
        "last_word_output_sample": 3667200
      },
      {
        "candidate_id": "de-conversation-05-podcast-segment-free-u11-c00",
        "output_offset_sample": 3682560,
        "first_word_output_sample": 3686400,
        "last_word_output_sample": 3978240
      },
      {
        "candidate_id": "de-conversation-05-podcast-segment-free-u12-c00",
        "output_offset_sample": 3993600,
        "first_word_output_sample": 3997440,
        "last_word_output_sample": 4300800
      },
      {
        "candidate_id": "de-conversation-05-podcast-segment-free-u13-c00",
        "output_offset_sample": 4316160,
        "first_word_output_sample": 4320000,
        "last_word_output_sample": 4627200
      },
      {
        "candidate_id": "de-conversation-05-podcast-segment-free-u14-c00",
        "output_offset_sample": 4642560,
        "first_word_output_sample": 4646400,
        "last_word_output_sample": 5038080
      },
      {
        "candidate_id": "de-conversation-05-podcast-segment-free-u15-c00",
        "output_offset_sample": 5057280,
        "first_word_output_sample": 5057280,
        "last_word_output_sample": 5379840
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 112.23,
    "samples": 5387040,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/de-conversation-05-podcast-segment-free-n1.mp3"
  },
  {
    "id": "de-conversation-05-podcast-segment-timed-n1",
    "case": "de-conversation-05",
    "model": "podcast",
    "strategy": "segment",
    "timing": "timed",
    "best_of": 1,
    "block_size": 1,
    "plain": "Eine Hörerin erlebt erst Englisch, dann Deutsch, Französisch, Spanisch und Niederländisch. Dieselbe Modellfamilie, andere Stimmen und in jeder Sprache ein eigener emotionaler Rhythmus? Das untersucht dieses Hörgrid, mit kurzen Dialogen und längeren Gesprächen. Ich würde ohne Referenz anfangen und zwei Gastgeber erfinden lassen. Gewagt, aber spannend. Danach drei, klar als Versuch gekennzeichnet, weil das Training auf zwei Sprecher ausgerichtet war. Ich wünsche mir Wärme, Überraschung und ein bisschen Neckerei, aber keine vorgelesenen Regieanweisungen. Die vollständigen Skripte und Seeds bleiben sichtbar, damit auch interessante Fehler nachvollziehbar sind. Gut. Ich bin noch skeptisch, aber die nächste Antwort möchte ich jetzt wirklich hören. Ich mag Gespräche, in denen noch Platz für eine echte, unerwartete Reaktion bleibt. Ja, manchmal erzählt dieses kleine Zögern mehr als der nächste völlig glatt formulierte Satz. Dann hören wir genau hin, auch an den Stellen, an denen die schöne Illusion kurz zerbricht. Einverstanden. Ich möchte Wärme und Persönlichkeit hören und trotzdem erkennen, wer gerade spricht. Stell dir eine kleine Wissenschaftssendung vor, bei der ehrliche Neugier wichtiger ist als große Versprechen. Und die zweite Person darf sagen: Moment, erklär mir das bitte noch einmal, ich habe den Gedanken noch nicht verstanden. Genau. Das Gespräch soll aufeinander reagieren, und nach jedem Stimmungswechsel sollen dieselben zwei Menschen wiederzuerkennen sein. Dann ist es ein fairer Vergleich. Wir lassen die Worte gleich und hören, was die verschiedenen Verfahren daraus machen.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/e29fee7effee7647585613560468e125c5806678de71e15b629fe25a408b3e74.mp3",
    "review_candidates": [
      "de-conversation-05-podcast-segment-timed-u01-c00",
      "de-conversation-05-podcast-segment-timed-u05-c00",
      "de-conversation-05-podcast-segment-timed-u12-c00"
    ]
  },
  {
    "id": "en-conversation-01-base17-block-free-n1",
    "case": "en-conversation-01",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 1,
    "block_size": 4,
    "plain": "I have been listening to Humaneness Podcast, a fine-tune of Humaneness Voice Large. You sound suspiciously excited. Did a speech model actually surprise you? It can read a script with different speakers and sentence-level acting directions. Wait, so I can ask one voice to doubt you while the other celebrates? Exactly. We are testing forty emotion labels, not claiming forty perfect performances. That distinction makes me trust the demonstration a little more. Annoyingly sensible of you. The new training mixes real conversations with acting examples to retain both formats. All right, show me the awkward pause before someone finally admits they are impressed. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/0df4a61c9919cf2fa5f7c61427b3e847e40d7d660afbace8425a3e5081555fd5.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-01-base17-block-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 833280
      },
      {
        "candidate_id": "en-conversation-01-base17-block-free-u01-c00",
        "output_offset_sample": 852480,
        "first_word_output_sample": 852480,
        "last_word_output_sample": 1777920
      },
      {
        "candidate_id": "en-conversation-01-base17-block-free-u02-c00",
        "output_offset_sample": 1793280,
        "first_word_output_sample": 1797120,
        "last_word_output_sample": 2634240
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 55.03,
    "samples": 2641440,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-01-base17-block-free-n1.mp3"
  },
  {
    "id": "en-conversation-01-base17-block-free-n4",
    "case": "en-conversation-01",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 4,
    "block_size": 4,
    "plain": "I have been listening to Humaneness Podcast, a fine-tune of Humaneness Voice Large. You sound suspiciously excited. Did a speech model actually surprise you? It can read a script with different speakers and sentence-level acting directions. Wait, so I can ask one voice to doubt you while the other celebrates? Exactly. We are testing forty emotion labels, not claiming forty perfect performances. That distinction makes me trust the demonstration a little more. Annoyingly sensible of you. The new training mixes real conversations with acting examples to retain both formats. All right, show me the awkward pause before someone finally admits they are impressed. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/f549bdf53f844ed83ec5182184855a62e43a1013b112c181ac9410f45dfd8db5.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-01-base17-block-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 833280
      },
      {
        "candidate_id": "en-conversation-01-base17-block-free-u01-c01",
        "output_offset_sample": 848640,
        "first_word_output_sample": 852480,
        "last_word_output_sample": 1839360
      },
      {
        "candidate_id": "en-conversation-01-base17-block-free-u02-c00",
        "output_offset_sample": 1854720,
        "first_word_output_sample": 1858560,
        "last_word_output_sample": 2695680
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 56.31,
    "samples": 2702880,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-01-base17-block-free-n4.mp3"
  },
  {
    "id": "en-conversation-01-base17-block-free-n8",
    "case": "en-conversation-01",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 8,
    "block_size": 4,
    "plain": "I have been listening to Humaneness Podcast, a fine-tune of Humaneness Voice Large. You sound suspiciously excited. Did a speech model actually surprise you? It can read a script with different speakers and sentence-level acting directions. Wait, so I can ask one voice to doubt you while the other celebrates? Exactly. We are testing forty emotion labels, not claiming forty perfect performances. That distinction makes me trust the demonstration a little more. Annoyingly sensible of you. The new training mixes real conversations with acting examples to retain both formats. All right, show me the awkward pause before someone finally admits they are impressed. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/7f3564b83d91a61257967eb5a8253f8b9df4039e78f721c3394f743fe8687b2d.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-01-base17-block-free-u00-c07",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 833280
      },
      {
        "candidate_id": "en-conversation-01-base17-block-free-u01-c06",
        "output_offset_sample": 848640,
        "first_word_output_sample": 852480,
        "last_word_output_sample": 1747200
      },
      {
        "candidate_id": "en-conversation-01-base17-block-free-u02-c00",
        "output_offset_sample": 1762560,
        "first_word_output_sample": 1766400,
        "last_word_output_sample": 2603520
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 54.39,
    "samples": 2610720,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-01-base17-block-free-n8.mp3"
  },
  {
    "id": "en-conversation-01-base17-block-timed-n1",
    "case": "en-conversation-01",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 1,
    "block_size": 4,
    "plain": "I have been listening to Humaneness Podcast, a fine-tune of Humaneness Voice Large. You sound suspiciously excited. Did a speech model actually surprise you? It can read a script with different speakers and sentence-level acting directions. Wait, so I can ask one voice to doubt you while the other celebrates? Exactly. We are testing forty emotion labels, not claiming forty perfect performances. That distinction makes me trust the demonstration a little more. Annoyingly sensible of you. The new training mixes real conversations with acting examples to retain both formats. All right, show me the awkward pause before someone finally admits they are impressed. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/13f6b88ce194140e8028c5dbc867a9f07c0dfbcf1f02cbacad8a78d27d8164ed.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-01-base17-block-timed-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 783360
      },
      {
        "candidate_id": "en-conversation-01-base17-block-timed-u01-c00",
        "output_offset_sample": 798720,
        "first_word_output_sample": 802560,
        "last_word_output_sample": 1639680
      },
      {
        "candidate_id": "en-conversation-01-base17-block-timed-u02-c00",
        "output_offset_sample": 1655040,
        "first_word_output_sample": 1658880,
        "last_word_output_sample": 2530560
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 52.87,
    "samples": 2537760,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-01-base17-block-timed-n1.mp3"
  },
  {
    "id": "en-conversation-01-base17-block-timed-n4",
    "case": "en-conversation-01",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 4,
    "block_size": 4,
    "plain": "I have been listening to Humaneness Podcast, a fine-tune of Humaneness Voice Large. You sound suspiciously excited. Did a speech model actually surprise you? It can read a script with different speakers and sentence-level acting directions. Wait, so I can ask one voice to doubt you while the other celebrates? Exactly. We are testing forty emotion labels, not claiming forty perfect performances. That distinction makes me trust the demonstration a little more. Annoyingly sensible of you. The new training mixes real conversations with acting examples to retain both formats. All right, show me the awkward pause before someone finally admits they are impressed. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/818ed0dd14e2ba210ba9d268f8919aee68b9c2b795e5883b49fbe157bcf99fdd.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-01-base17-block-timed-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 783360
      },
      {
        "candidate_id": "en-conversation-01-base17-block-timed-u01-c01",
        "output_offset_sample": 798720,
        "first_word_output_sample": 802560,
        "last_word_output_sample": 1639680
      },
      {
        "candidate_id": "en-conversation-01-base17-block-timed-u02-c01",
        "output_offset_sample": 1655040,
        "first_word_output_sample": 1658880,
        "last_word_output_sample": 2530560
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 52.87,
    "samples": 2537760,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-01-base17-block-timed-n4.mp3"
  },
  {
    "id": "en-conversation-01-base17-block-timed-n8",
    "case": "en-conversation-01",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 8,
    "block_size": 4,
    "plain": "I have been listening to Humaneness Podcast, a fine-tune of Humaneness Voice Large. You sound suspiciously excited. Did a speech model actually surprise you? It can read a script with different speakers and sentence-level acting directions. Wait, so I can ask one voice to doubt you while the other celebrates? Exactly. We are testing forty emotion labels, not claiming forty perfect performances. That distinction makes me trust the demonstration a little more. Annoyingly sensible of you. The new training mixes real conversations with acting examples to retain both formats. All right, show me the awkward pause before someone finally admits they are impressed. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/f0f14a034e8b435e793b5f76b85b532d40d3ff8d8b8e047ca111f0a83bb02ace.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-01-base17-block-timed-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 783360
      },
      {
        "candidate_id": "en-conversation-01-base17-block-timed-u01-c01",
        "output_offset_sample": 798720,
        "first_word_output_sample": 802560,
        "last_word_output_sample": 1639680
      },
      {
        "candidate_id": "en-conversation-01-base17-block-timed-u02-c04",
        "output_offset_sample": 1655040,
        "first_word_output_sample": 1658880,
        "last_word_output_sample": 2530560
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 52.87,
    "samples": 2537760,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-01-base17-block-timed-n8.mp3"
  },
  {
    "id": "en-conversation-01-base17-segment-free-n1",
    "case": "en-conversation-01",
    "model": "base17",
    "strategy": "segment",
    "timing": "free",
    "best_of": 1,
    "block_size": 1,
    "plain": "I have been listening to Humaneness Podcast, a fine-tune of Humaneness Voice Large. You sound suspiciously excited. Did a speech model actually surprise you? It can read a script with different speakers and sentence-level acting directions. Wait, so I can ask one voice to doubt you while the other celebrates? Exactly. We are testing forty emotion labels, not claiming forty perfect performances. That distinction makes me trust the demonstration a little more. Annoyingly sensible of you. The new training mixes real conversations with acting examples to retain both formats. All right, show me the awkward pause before someone finally admits they are impressed. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/bfc133e3f97dae6a039072b8c2b728986441dbdd1d1388255ced7112a11ba485.mp3",
    "review_candidates": [
      "en-conversation-01-base17-segment-free-u07-c00"
    ]
  },
  {
    "id": "en-conversation-01-base17-segment-timed-n1",
    "case": "en-conversation-01",
    "model": "base17",
    "strategy": "segment",
    "timing": "timed",
    "best_of": 1,
    "block_size": 1,
    "plain": "I have been listening to Humaneness Podcast, a fine-tune of Humaneness Voice Large. You sound suspiciously excited. Did a speech model actually surprise you? It can read a script with different speakers and sentence-level acting directions. Wait, so I can ask one voice to doubt you while the other celebrates? Exactly. We are testing forty emotion labels, not claiming forty perfect performances. That distinction makes me trust the demonstration a little more. Annoyingly sensible of you. The new training mixes real conversations with acting examples to retain both formats. All right, show me the awkward pause before someone finally admits they are impressed. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/bb2e7ffa4ac0a492d10ccebb5ad164f4b9dd9e2370e2a371919e969e40caf4ac.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-01-base17-segment-timed-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 195840
      },
      {
        "candidate_id": "en-conversation-01-base17-segment-timed-u01-c00",
        "output_offset_sample": 215040,
        "first_word_output_sample": 215040,
        "last_word_output_sample": 387840
      },
      {
        "candidate_id": "en-conversation-01-base17-segment-timed-u02-c00",
        "output_offset_sample": 407040,
        "first_word_output_sample": 407040,
        "last_word_output_sample": 599040
      },
      {
        "candidate_id": "en-conversation-01-base17-segment-timed-u03-c00",
        "output_offset_sample": 614400,
        "first_word_output_sample": 618240,
        "last_word_output_sample": 825600
      },
      {
        "candidate_id": "en-conversation-01-base17-segment-timed-u04-c00",
        "output_offset_sample": 840960,
        "first_word_output_sample": 844800,
        "last_word_output_sample": 1025280
      },
      {
        "candidate_id": "en-conversation-01-base17-segment-timed-u05-c00",
        "output_offset_sample": 1044480,
        "first_word_output_sample": 1044480,
        "last_word_output_sample": 1259520
      },
      {
        "candidate_id": "en-conversation-01-base17-segment-timed-u06-c00",
        "output_offset_sample": 1278720,
        "first_word_output_sample": 1278720,
        "last_word_output_sample": 1482240
      },
      {
        "candidate_id": "en-conversation-01-base17-segment-timed-u07-c00",
        "output_offset_sample": 1497600,
        "first_word_output_sample": 1501440,
        "last_word_output_sample": 1712640
      },
      {
        "candidate_id": "en-conversation-01-base17-segment-timed-u08-c00",
        "output_offset_sample": 1728000,
        "first_word_output_sample": 1731840,
        "last_word_output_sample": 1943040
      },
      {
        "candidate_id": "en-conversation-01-base17-segment-timed-u09-c00",
        "output_offset_sample": 1958400,
        "first_word_output_sample": 1962240,
        "last_word_output_sample": 2142720
      },
      {
        "candidate_id": "en-conversation-01-base17-segment-timed-u10-c00",
        "output_offset_sample": 2161920,
        "first_word_output_sample": 2161920,
        "last_word_output_sample": 2376960
      },
      {
        "candidate_id": "en-conversation-01-base17-segment-timed-u11-c00",
        "output_offset_sample": 2392320,
        "first_word_output_sample": 2396160,
        "last_word_output_sample": 2618880
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 54.71,
    "samples": 2626080,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-01-base17-segment-timed-n1.mp3"
  },
  {
    "id": "en-conversation-01-podcast-block-free-n1",
    "case": "en-conversation-01",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 1,
    "block_size": 4,
    "plain": "I have been listening to Humaneness Podcast, a fine-tune of Humaneness Voice Large. You sound suspiciously excited. Did a speech model actually surprise you? It can read a script with different speakers and sentence-level acting directions. Wait, so I can ask one voice to doubt you while the other celebrates? Exactly. We are testing forty emotion labels, not claiming forty perfect performances. That distinction makes me trust the demonstration a little more. Annoyingly sensible of you. The new training mixes real conversations with acting examples to retain both formats. All right, show me the awkward pause before someone finally admits they are impressed. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/6b44de43563fb8e489bd3cd550076bfca931edf6ac0b6051f0228d6fbe3fc16c.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-01-podcast-block-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 825600
      },
      {
        "candidate_id": "en-conversation-01-podcast-block-free-u01-c00",
        "output_offset_sample": 844800,
        "first_word_output_sample": 844800,
        "last_word_output_sample": 1808640
      },
      {
        "candidate_id": "en-conversation-01-podcast-block-free-u02-c00",
        "output_offset_sample": 1824000,
        "first_word_output_sample": 1827840,
        "last_word_output_sample": 2711040
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 56.63,
    "samples": 2718240,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-01-podcast-block-free-n1.mp3"
  },
  {
    "id": "en-conversation-01-podcast-block-free-n4",
    "case": "en-conversation-01",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 4,
    "block_size": 4,
    "plain": "I have been listening to Humaneness Podcast, a fine-tune of Humaneness Voice Large. You sound suspiciously excited. Did a speech model actually surprise you? It can read a script with different speakers and sentence-level acting directions. Wait, so I can ask one voice to doubt you while the other celebrates? Exactly. We are testing forty emotion labels, not claiming forty perfect performances. That distinction makes me trust the demonstration a little more. Annoyingly sensible of you. The new training mixes real conversations with acting examples to retain both formats. All right, show me the awkward pause before someone finally admits they are impressed. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/29f8d222568fa872e1d4fecf26331f3d739a61fee099f80f35c49833c921bb1d.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-01-podcast-block-free-u00-c01",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 833280
      },
      {
        "candidate_id": "en-conversation-01-podcast-block-free-u01-c02",
        "output_offset_sample": 848640,
        "first_word_output_sample": 852480,
        "last_word_output_sample": 1777920
      },
      {
        "candidate_id": "en-conversation-01-podcast-block-free-u02-c03",
        "output_offset_sample": 1797120,
        "first_word_output_sample": 1797120,
        "last_word_output_sample": 2718720
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 56.79,
    "samples": 2725920,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-01-podcast-block-free-n4.mp3"
  },
  {
    "id": "en-conversation-01-podcast-block-free-n8",
    "case": "en-conversation-01",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 8,
    "block_size": 4,
    "plain": "I have been listening to Humaneness Podcast, a fine-tune of Humaneness Voice Large. You sound suspiciously excited. Did a speech model actually surprise you? It can read a script with different speakers and sentence-level acting directions. Wait, so I can ask one voice to doubt you while the other celebrates? Exactly. We are testing forty emotion labels, not claiming forty perfect performances. That distinction makes me trust the demonstration a little more. Annoyingly sensible of you. The new training mixes real conversations with acting examples to retain both formats. All right, show me the awkward pause before someone finally admits they are impressed. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/88fee28db93cf3d3f96272945ee4902687d95517de336c873d592a1d8a3aca70.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-01-podcast-block-free-u00-c01",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 833280
      },
      {
        "candidate_id": "en-conversation-01-podcast-block-free-u01-c05",
        "output_offset_sample": 852480,
        "first_word_output_sample": 852480,
        "last_word_output_sample": 1785600
      },
      {
        "candidate_id": "en-conversation-01-podcast-block-free-u02-c03",
        "output_offset_sample": 1804800,
        "first_word_output_sample": 1804800,
        "last_word_output_sample": 2726400
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 56.95,
    "samples": 2733600,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-01-podcast-block-free-n8.mp3"
  },
  {
    "id": "en-conversation-01-podcast-block-timed-n1",
    "case": "en-conversation-01",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 1,
    "block_size": 4,
    "plain": "I have been listening to Humaneness Podcast, a fine-tune of Humaneness Voice Large. You sound suspiciously excited. Did a speech model actually surprise you? It can read a script with different speakers and sentence-level acting directions. Wait, so I can ask one voice to doubt you while the other celebrates? Exactly. We are testing forty emotion labels, not claiming forty perfect performances. That distinction makes me trust the demonstration a little more. Annoyingly sensible of you. The new training mixes real conversations with acting examples to retain both formats. All right, show me the awkward pause before someone finally admits they are impressed. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/78ed755afcb516e5f3daad8fda60f53c33d0147c421ee4c8d905afc5380dd9e4.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-01-podcast-block-timed-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 787200
      },
      {
        "candidate_id": "en-conversation-01-podcast-block-timed-u01-c00",
        "output_offset_sample": 806400,
        "first_word_output_sample": 806400,
        "last_word_output_sample": 1651200
      },
      {
        "candidate_id": "en-conversation-01-podcast-block-timed-u02-c00",
        "output_offset_sample": 1670400,
        "first_word_output_sample": 1670400,
        "last_word_output_sample": 2545920
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 53.19,
    "samples": 2553120,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-01-podcast-block-timed-n1.mp3"
  },
  {
    "id": "en-conversation-01-podcast-block-timed-n4",
    "case": "en-conversation-01",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 4,
    "block_size": 4,
    "plain": "I have been listening to Humaneness Podcast, a fine-tune of Humaneness Voice Large. You sound suspiciously excited. Did a speech model actually surprise you? It can read a script with different speakers and sentence-level acting directions. Wait, so I can ask one voice to doubt you while the other celebrates? Exactly. We are testing forty emotion labels, not claiming forty perfect performances. That distinction makes me trust the demonstration a little more. Annoyingly sensible of you. The new training mixes real conversations with acting examples to retain both formats. All right, show me the awkward pause before someone finally admits they are impressed. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/3de55dec840be480f59f4e259fb983b3dbe763cbbcb7db257a6f3e66580c0d45.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-01-podcast-block-timed-u00-c01",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 787200
      },
      {
        "candidate_id": "en-conversation-01-podcast-block-timed-u01-c00",
        "output_offset_sample": 806400,
        "first_word_output_sample": 806400,
        "last_word_output_sample": 1651200
      },
      {
        "candidate_id": "en-conversation-01-podcast-block-timed-u02-c02",
        "output_offset_sample": 1666560,
        "first_word_output_sample": 1670400,
        "last_word_output_sample": 2534400
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 52.95,
    "samples": 2541600,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-01-podcast-block-timed-n4.mp3"
  },
  {
    "id": "en-conversation-01-podcast-block-timed-n8",
    "case": "en-conversation-01",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 8,
    "block_size": 4,
    "plain": "I have been listening to Humaneness Podcast, a fine-tune of Humaneness Voice Large. You sound suspiciously excited. Did a speech model actually surprise you? It can read a script with different speakers and sentence-level acting directions. Wait, so I can ask one voice to doubt you while the other celebrates? Exactly. We are testing forty emotion labels, not claiming forty perfect performances. That distinction makes me trust the demonstration a little more. Annoyingly sensible of you. The new training mixes real conversations with acting examples to retain both formats. All right, show me the awkward pause before someone finally admits they are impressed. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/3de55dec840be480f59f4e259fb983b3dbe763cbbcb7db257a6f3e66580c0d45.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-01-podcast-block-timed-u00-c01",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 787200
      },
      {
        "candidate_id": "en-conversation-01-podcast-block-timed-u01-c00",
        "output_offset_sample": 806400,
        "first_word_output_sample": 806400,
        "last_word_output_sample": 1651200
      },
      {
        "candidate_id": "en-conversation-01-podcast-block-timed-u02-c02",
        "output_offset_sample": 1666560,
        "first_word_output_sample": 1670400,
        "last_word_output_sample": 2534400
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 52.95,
    "samples": 2541600,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-01-podcast-block-timed-n8.mp3"
  },
  {
    "id": "en-conversation-01-podcast-segment-free-n1",
    "case": "en-conversation-01",
    "model": "podcast",
    "strategy": "segment",
    "timing": "free",
    "best_of": 1,
    "block_size": 1,
    "plain": "I have been listening to Humaneness Podcast, a fine-tune of Humaneness Voice Large. You sound suspiciously excited. Did a speech model actually surprise you? It can read a script with different speakers and sentence-level acting directions. Wait, so I can ask one voice to doubt you while the other celebrates? Exactly. We are testing forty emotion labels, not claiming forty perfect performances. That distinction makes me trust the demonstration a little more. Annoyingly sensible of you. The new training mixes real conversations with acting examples to retain both formats. All right, show me the awkward pause before someone finally admits they are impressed. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": null,
    "review_candidates": [
      "en-conversation-01-podcast-segment-free-u07-c00"
    ]
  },
  {
    "id": "en-conversation-01-podcast-segment-timed-n1",
    "case": "en-conversation-01",
    "model": "podcast",
    "strategy": "segment",
    "timing": "timed",
    "best_of": 1,
    "block_size": 1,
    "plain": "I have been listening to Humaneness Podcast, a fine-tune of Humaneness Voice Large. You sound suspiciously excited. Did a speech model actually surprise you? It can read a script with different speakers and sentence-level acting directions. Wait, so I can ask one voice to doubt you while the other celebrates? Exactly. We are testing forty emotion labels, not claiming forty perfect performances. That distinction makes me trust the demonstration a little more. Annoyingly sensible of you. The new training mixes real conversations with acting examples to retain both formats. All right, show me the awkward pause before someone finally admits they are impressed. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/4093464dedcceea22955f7618cfb6f78a395506d0ad6d9f32a30fc463195db88.mp3",
    "review_candidates": [
      "en-conversation-01-podcast-segment-timed-u00-c00",
      "en-conversation-01-podcast-segment-timed-u02-c00"
    ]
  },
  {
    "id": "en-conversation-02-base17-block-free-n1",
    "case": "en-conversation-02",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 1,
    "block_size": 5,
    "plain": "Imagine a tiny radio studio where the hosts have just discovered this voice model. Please tell me the tiny studio has a very dramatic red recording light. Of course. Humaneness Voice Large supplies the foundation; podcast fine-tuning teaches dialogue. And does the second host interrupt with healthy skepticism, or just politely agree? Healthy skepticism, please. We give each turn its own mood and its own words. I refuse to be dazzled by a cheerful description. Let me hear the voices change. That is why the reference clips sit beside the output. You can compare for yourself. Now that is a demonstration I can enjoy. Keep the little red light on. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/7d0848507ac7f044a4df866b05bffa717680e7f889a7bb65c09d6bb0dc80dd83.mp3",
    "review_candidates": [
      "en-conversation-02-base17-block-free-u00-c00",
      "en-conversation-02-base17-block-free-u01-c00"
    ]
  },
  {
    "id": "en-conversation-02-base17-block-free-n4",
    "case": "en-conversation-02",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 4,
    "block_size": 5,
    "plain": "Imagine a tiny radio studio where the hosts have just discovered this voice model. Please tell me the tiny studio has a very dramatic red recording light. Of course. Humaneness Voice Large supplies the foundation; podcast fine-tuning teaches dialogue. And does the second host interrupt with healthy skepticism, or just politely agree? Healthy skepticism, please. We give each turn its own mood and its own words. I refuse to be dazzled by a cheerful description. Let me hear the voices change. That is why the reference clips sit beside the output. You can compare for yourself. Now that is a demonstration I can enjoy. Keep the little red light on. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/9e0d6e3e9e2ee2c4f75c0f06f38d7eaaef527693f412900c73f9108bcca82e2c.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-02-base17-block-free-u00-c01",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 1048320
      },
      {
        "candidate_id": "en-conversation-02-base17-block-free-u01-c02",
        "output_offset_sample": 1063680,
        "first_word_output_sample": 1067520,
        "last_word_output_sample": 2060160
      },
      {
        "candidate_id": "en-conversation-02-base17-block-free-u02-c01",
        "output_offset_sample": 2079360,
        "first_word_output_sample": 2079360,
        "last_word_output_sample": 2567040
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 53.63,
    "samples": 2574240,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-02-base17-block-free-n4.mp3"
  },
  {
    "id": "en-conversation-02-base17-block-free-n8",
    "case": "en-conversation-02",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 8,
    "block_size": 5,
    "plain": "Imagine a tiny radio studio where the hosts have just discovered this voice model. Please tell me the tiny studio has a very dramatic red recording light. Of course. Humaneness Voice Large supplies the foundation; podcast fine-tuning teaches dialogue. And does the second host interrupt with healthy skepticism, or just politely agree? Healthy skepticism, please. We give each turn its own mood and its own words. I refuse to be dazzled by a cheerful description. Let me hear the voices change. That is why the reference clips sit beside the output. You can compare for yourself. Now that is a demonstration I can enjoy. Keep the little red light on. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/144e0d9f71851d91f6fdfdf2c1ed2183a629c95f6614e88896892b286ba43bab.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-02-base17-block-free-u00-c01",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 1048320
      },
      {
        "candidate_id": "en-conversation-02-base17-block-free-u01-c06",
        "output_offset_sample": 1063680,
        "first_word_output_sample": 1067520,
        "last_word_output_sample": 2088960
      },
      {
        "candidate_id": "en-conversation-02-base17-block-free-u02-c06",
        "output_offset_sample": 2104320,
        "first_word_output_sample": 2108160,
        "last_word_output_sample": 2576640
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 53.83,
    "samples": 2583840,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-02-base17-block-free-n8.mp3"
  },
  {
    "id": "en-conversation-02-base17-block-timed-n1",
    "case": "en-conversation-02",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 1,
    "block_size": 5,
    "plain": "Imagine a tiny radio studio where the hosts have just discovered this voice model. Please tell me the tiny studio has a very dramatic red recording light. Of course. Humaneness Voice Large supplies the foundation; podcast fine-tuning teaches dialogue. And does the second host interrupt with healthy skepticism, or just politely agree? Healthy skepticism, please. We give each turn its own mood and its own words. I refuse to be dazzled by a cheerful description. Let me hear the voices change. That is why the reference clips sit beside the output. You can compare for yourself. Now that is a demonstration I can enjoy. Keep the little red light on. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/4eef8bc7272a693cf0f8b48ca8f68051e23b6a5d318c5bbc3d11e62e48e72e90.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-02-base17-block-timed-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 1048320
      },
      {
        "candidate_id": "en-conversation-02-base17-block-timed-u01-c00",
        "output_offset_sample": 1063680,
        "first_word_output_sample": 1067520,
        "last_word_output_sample": 2169600
      },
      {
        "candidate_id": "en-conversation-02-base17-block-timed-u02-c00",
        "output_offset_sample": 2184960,
        "first_word_output_sample": 2188800,
        "last_word_output_sample": 2638080
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 55.11,
    "samples": 2645280,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-02-base17-block-timed-n1.mp3"
  },
  {
    "id": "en-conversation-02-base17-block-timed-n4",
    "case": "en-conversation-02",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 4,
    "block_size": 5,
    "plain": "Imagine a tiny radio studio where the hosts have just discovered this voice model. Please tell me the tiny studio has a very dramatic red recording light. Of course. Humaneness Voice Large supplies the foundation; podcast fine-tuning teaches dialogue. And does the second host interrupt with healthy skepticism, or just politely agree? Healthy skepticism, please. We give each turn its own mood and its own words. I refuse to be dazzled by a cheerful description. Let me hear the voices change. That is why the reference clips sit beside the output. You can compare for yourself. Now that is a demonstration I can enjoy. Keep the little red light on. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/59b8d3ec28b9049cc51ce759a3f63d998b86bc1c605a6cd524fa00e33b93e7b5.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-02-base17-block-timed-u00-c03",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 1048320
      },
      {
        "candidate_id": "en-conversation-02-base17-block-timed-u01-c03",
        "output_offset_sample": 1063680,
        "first_word_output_sample": 1067520,
        "last_word_output_sample": 2173440
      },
      {
        "candidate_id": "en-conversation-02-base17-block-timed-u02-c01",
        "output_offset_sample": 2188800,
        "first_word_output_sample": 2192640,
        "last_word_output_sample": 2645760
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 55.27,
    "samples": 2652960,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-02-base17-block-timed-n4.mp3"
  },
  {
    "id": "en-conversation-02-base17-block-timed-n8",
    "case": "en-conversation-02",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 8,
    "block_size": 5,
    "plain": "Imagine a tiny radio studio where the hosts have just discovered this voice model. Please tell me the tiny studio has a very dramatic red recording light. Of course. Humaneness Voice Large supplies the foundation; podcast fine-tuning teaches dialogue. And does the second host interrupt with healthy skepticism, or just politely agree? Healthy skepticism, please. We give each turn its own mood and its own words. I refuse to be dazzled by a cheerful description. Let me hear the voices change. That is why the reference clips sit beside the output. You can compare for yourself. Now that is a demonstration I can enjoy. Keep the little red light on. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/f25f027b1a7d7018233011b72b66c1bf5f8d5e68ae2152c2568f8a8231282aa5.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-02-base17-block-timed-u00-c03",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 1048320
      },
      {
        "candidate_id": "en-conversation-02-base17-block-timed-u01-c05",
        "output_offset_sample": 1063680,
        "first_word_output_sample": 1067520,
        "last_word_output_sample": 2173440
      },
      {
        "candidate_id": "en-conversation-02-base17-block-timed-u02-c04",
        "output_offset_sample": 2188800,
        "first_word_output_sample": 2192640,
        "last_word_output_sample": 2645760
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 55.27,
    "samples": 2652960,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-02-base17-block-timed-n8.mp3"
  },
  {
    "id": "en-conversation-02-base17-segment-free-n1",
    "case": "en-conversation-02",
    "model": "base17",
    "strategy": "segment",
    "timing": "free",
    "best_of": 1,
    "block_size": 1,
    "plain": "Imagine a tiny radio studio where the hosts have just discovered this voice model. Please tell me the tiny studio has a very dramatic red recording light. Of course. Humaneness Voice Large supplies the foundation; podcast fine-tuning teaches dialogue. And does the second host interrupt with healthy skepticism, or just politely agree? Healthy skepticism, please. We give each turn its own mood and its own words. I refuse to be dazzled by a cheerful description. Let me hear the voices change. That is why the reference clips sit beside the output. You can compare for yourself. Now that is a demonstration I can enjoy. Keep the little red light on. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/06bc888edb10ef4dfdd1509034c122a69675c4e1e993151409002e441ea373ef.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-02-base17-segment-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 268800
      },
      {
        "candidate_id": "en-conversation-02-base17-segment-free-u01-c00",
        "output_offset_sample": 284160,
        "first_word_output_sample": 288000,
        "last_word_output_sample": 514560
      },
      {
        "candidate_id": "en-conversation-02-base17-segment-free-u02-c00",
        "output_offset_sample": 533760,
        "first_word_output_sample": 533760,
        "last_word_output_sample": 1056000
      },
      {
        "candidate_id": "en-conversation-02-base17-segment-free-u03-c00",
        "output_offset_sample": 1071360,
        "first_word_output_sample": 1075200,
        "last_word_output_sample": 1351680
      },
      {
        "candidate_id": "en-conversation-02-base17-segment-free-u04-c00",
        "output_offset_sample": 1367040,
        "first_word_output_sample": 1370880,
        "last_word_output_sample": 1785600
      },
      {
        "candidate_id": "en-conversation-02-base17-segment-free-u05-c00",
        "output_offset_sample": 1800960,
        "first_word_output_sample": 1804800,
        "last_word_output_sample": 2277120
      },
      {
        "candidate_id": "en-conversation-02-base17-segment-free-u06-c00",
        "output_offset_sample": 2296320,
        "first_word_output_sample": 2296320,
        "last_word_output_sample": 2542080
      },
      {
        "candidate_id": "en-conversation-02-base17-segment-free-u07-c00",
        "output_offset_sample": 2561280,
        "first_word_output_sample": 2561280,
        "last_word_output_sample": 2791680
      },
      {
        "candidate_id": "en-conversation-02-base17-segment-free-u08-c00",
        "output_offset_sample": 2807040,
        "first_word_output_sample": 2810880,
        "last_word_output_sample": 3137280
      },
      {
        "candidate_id": "en-conversation-02-base17-segment-free-u09-c00",
        "output_offset_sample": 3152640,
        "first_word_output_sample": 3156480,
        "last_word_output_sample": 3598080
      },
      {
        "candidate_id": "en-conversation-02-base17-segment-free-u10-c00",
        "output_offset_sample": 3617280,
        "first_word_output_sample": 3617280,
        "last_word_output_sample": 3866880
      },
      {
        "candidate_id": "en-conversation-02-base17-segment-free-u11-c00",
        "output_offset_sample": 3882240,
        "first_word_output_sample": 3886080,
        "last_word_output_sample": 4281600
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 89.35,
    "samples": 4288800,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-02-base17-segment-free-n1.mp3"
  },
  {
    "id": "en-conversation-02-base17-segment-timed-n1",
    "case": "en-conversation-02",
    "model": "base17",
    "strategy": "segment",
    "timing": "timed",
    "best_of": 1,
    "block_size": 1,
    "plain": "Imagine a tiny radio studio where the hosts have just discovered this voice model. Please tell me the tiny studio has a very dramatic red recording light. Of course. Humaneness Voice Large supplies the foundation; podcast fine-tuning teaches dialogue. And does the second host interrupt with healthy skepticism, or just politely agree? Healthy skepticism, please. We give each turn its own mood and its own words. I refuse to be dazzled by a cheerful description. Let me hear the voices change. That is why the reference clips sit beside the output. You can compare for yourself. Now that is a demonstration I can enjoy. Keep the little red light on. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/d6db508bb12ee9c79412595bcf187d8d6afee046c97b4d071f81622f3be8a9dc.mp3",
    "review_candidates": [
      "en-conversation-02-base17-segment-timed-u02-c00"
    ]
  },
  {
    "id": "en-conversation-02-podcast-block-free-n1",
    "case": "en-conversation-02",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 1,
    "block_size": 5,
    "plain": "Imagine a tiny radio studio where the hosts have just discovered this voice model. Please tell me the tiny studio has a very dramatic red recording light. Of course. Humaneness Voice Large supplies the foundation; podcast fine-tuning teaches dialogue. And does the second host interrupt with healthy skepticism, or just politely agree? Healthy skepticism, please. We give each turn its own mood and its own words. I refuse to be dazzled by a cheerful description. Let me hear the voices change. That is why the reference clips sit beside the output. You can compare for yourself. Now that is a demonstration I can enjoy. Keep the little red light on. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/4d7967db62cfede4eb1901b5844f06cc641b7a44e658821ad87916d014ad35d8.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-02-podcast-block-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 1102080
      },
      {
        "candidate_id": "en-conversation-02-podcast-block-free-u01-c00",
        "output_offset_sample": 1117440,
        "first_word_output_sample": 1121280,
        "last_word_output_sample": 2223360
      },
      {
        "candidate_id": "en-conversation-02-podcast-block-free-u02-c00",
        "output_offset_sample": 2242560,
        "first_word_output_sample": 2242560,
        "last_word_output_sample": 2726400
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 56.95,
    "samples": 2733600,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-02-podcast-block-free-n1.mp3"
  },
  {
    "id": "en-conversation-02-podcast-block-free-n4",
    "case": "en-conversation-02",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 4,
    "block_size": 5,
    "plain": "Imagine a tiny radio studio where the hosts have just discovered this voice model. Please tell me the tiny studio has a very dramatic red recording light. Of course. Humaneness Voice Large supplies the foundation; podcast fine-tuning teaches dialogue. And does the second host interrupt with healthy skepticism, or just politely agree? Healthy skepticism, please. We give each turn its own mood and its own words. I refuse to be dazzled by a cheerful description. Let me hear the voices change. That is why the reference clips sit beside the output. You can compare for yourself. Now that is a demonstration I can enjoy. Keep the little red light on. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/9938418d19d3b1eb2786aad7a78562d8eec5a122154b19693e7a9de0a0e8f9f4.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-02-podcast-block-free-u00-c02",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 1063680
      },
      {
        "candidate_id": "en-conversation-02-podcast-block-free-u01-c01",
        "output_offset_sample": 1082880,
        "first_word_output_sample": 1082880,
        "last_word_output_sample": 2196480
      },
      {
        "candidate_id": "en-conversation-02-podcast-block-free-u02-c01",
        "output_offset_sample": 2215680,
        "first_word_output_sample": 2215680,
        "last_word_output_sample": 2711040
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 56.63,
    "samples": 2718240,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-02-podcast-block-free-n4.mp3"
  },
  {
    "id": "en-conversation-02-podcast-block-free-n8",
    "case": "en-conversation-02",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 8,
    "block_size": 5,
    "plain": "Imagine a tiny radio studio where the hosts have just discovered this voice model. Please tell me the tiny studio has a very dramatic red recording light. Of course. Humaneness Voice Large supplies the foundation; podcast fine-tuning teaches dialogue. And does the second host interrupt with healthy skepticism, or just politely agree? Healthy skepticism, please. We give each turn its own mood and its own words. I refuse to be dazzled by a cheerful description. Let me hear the voices change. That is why the reference clips sit beside the output. You can compare for yourself. Now that is a demonstration I can enjoy. Keep the little red light on. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/238c6732aa27a8a6076f483781ab68044355aeae3a4d240f552fa484bff529e2.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-02-podcast-block-free-u00-c07",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 1109760
      },
      {
        "candidate_id": "en-conversation-02-podcast-block-free-u01-c01",
        "output_offset_sample": 1128960,
        "first_word_output_sample": 1128960,
        "last_word_output_sample": 2242560
      },
      {
        "candidate_id": "en-conversation-02-podcast-block-free-u02-c06",
        "output_offset_sample": 2257920,
        "first_word_output_sample": 2261760,
        "last_word_output_sample": 2726400
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 56.95,
    "samples": 2733600,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-02-podcast-block-free-n8.mp3"
  },
  {
    "id": "en-conversation-02-podcast-block-timed-n1",
    "case": "en-conversation-02",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 1,
    "block_size": 5,
    "plain": "Imagine a tiny radio studio where the hosts have just discovered this voice model. Please tell me the tiny studio has a very dramatic red recording light. Of course. Humaneness Voice Large supplies the foundation; podcast fine-tuning teaches dialogue. And does the second host interrupt with healthy skepticism, or just politely agree? Healthy skepticism, please. We give each turn its own mood and its own words. I refuse to be dazzled by a cheerful description. Let me hear the voices change. That is why the reference clips sit beside the output. You can compare for yourself. Now that is a demonstration I can enjoy. Keep the little red light on. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/1909a407c7b8cba5a9165844ee9b09da1bd75815672d9397072b6a3b4d49bd2a.mp3",
    "review_candidates": [
      "en-conversation-02-podcast-block-timed-u00-c00"
    ]
  },
  {
    "id": "en-conversation-02-podcast-block-timed-n4",
    "case": "en-conversation-02",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 4,
    "block_size": 5,
    "plain": "Imagine a tiny radio studio where the hosts have just discovered this voice model. Please tell me the tiny studio has a very dramatic red recording light. Of course. Humaneness Voice Large supplies the foundation; podcast fine-tuning teaches dialogue. And does the second host interrupt with healthy skepticism, or just politely agree? Healthy skepticism, please. We give each turn its own mood and its own words. I refuse to be dazzled by a cheerful description. Let me hear the voices change. That is why the reference clips sit beside the output. You can compare for yourself. Now that is a demonstration I can enjoy. Keep the little red light on. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/69a7fe5fedcf9ac12482545213a04f42c5be91a1617cae8c72c92f5d291c5c28.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-02-podcast-block-timed-u00-c02",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 1052160
      },
      {
        "candidate_id": "en-conversation-02-podcast-block-timed-u01-c02",
        "output_offset_sample": 1071360,
        "first_word_output_sample": 1071360,
        "last_word_output_sample": 2184960
      },
      {
        "candidate_id": "en-conversation-02-podcast-block-timed-u02-c03",
        "output_offset_sample": 2204160,
        "first_word_output_sample": 2204160,
        "last_word_output_sample": 2661120
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 55.59,
    "samples": 2668320,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-02-podcast-block-timed-n4.mp3"
  },
  {
    "id": "en-conversation-02-podcast-block-timed-n8",
    "case": "en-conversation-02",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 8,
    "block_size": 5,
    "plain": "Imagine a tiny radio studio where the hosts have just discovered this voice model. Please tell me the tiny studio has a very dramatic red recording light. Of course. Humaneness Voice Large supplies the foundation; podcast fine-tuning teaches dialogue. And does the second host interrupt with healthy skepticism, or just politely agree? Healthy skepticism, please. We give each turn its own mood and its own words. I refuse to be dazzled by a cheerful description. Let me hear the voices change. That is why the reference clips sit beside the output. You can compare for yourself. Now that is a demonstration I can enjoy. Keep the little red light on. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/cedda632f67d0af61d6d5e4b947ce106e927e91c49a2f93e219081475f67772a.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-02-podcast-block-timed-u00-c05",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 1048320
      },
      {
        "candidate_id": "en-conversation-02-podcast-block-timed-u01-c07",
        "output_offset_sample": 1063680,
        "first_word_output_sample": 1067520,
        "last_word_output_sample": 2173440
      },
      {
        "candidate_id": "en-conversation-02-podcast-block-timed-u02-c06",
        "output_offset_sample": 2192640,
        "first_word_output_sample": 2192640,
        "last_word_output_sample": 2645760
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 55.27,
    "samples": 2652960,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-02-podcast-block-timed-n8.mp3"
  },
  {
    "id": "en-conversation-02-podcast-segment-free-n1",
    "case": "en-conversation-02",
    "model": "podcast",
    "strategy": "segment",
    "timing": "free",
    "best_of": 1,
    "block_size": 1,
    "plain": "Imagine a tiny radio studio where the hosts have just discovered this voice model. Please tell me the tiny studio has a very dramatic red recording light. Of course. Humaneness Voice Large supplies the foundation; podcast fine-tuning teaches dialogue. And does the second host interrupt with healthy skepticism, or just politely agree? Healthy skepticism, please. We give each turn its own mood and its own words. I refuse to be dazzled by a cheerful description. Let me hear the voices change. That is why the reference clips sit beside the output. You can compare for yourself. Now that is a demonstration I can enjoy. Keep the little red light on. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/198eb96b0324d5081b7ad122177b863ea2e88905482c96b64df291b603b3b466.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-02-podcast-segment-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 311040
      },
      {
        "candidate_id": "en-conversation-02-podcast-segment-free-u01-c00",
        "output_offset_sample": 326400,
        "first_word_output_sample": 330240,
        "last_word_output_sample": 591360
      },
      {
        "candidate_id": "en-conversation-02-podcast-segment-free-u02-c00",
        "output_offset_sample": 606720,
        "first_word_output_sample": 610560,
        "last_word_output_sample": 894720
      },
      {
        "candidate_id": "en-conversation-02-podcast-segment-free-u03-c00",
        "output_offset_sample": 910080,
        "first_word_output_sample": 913920,
        "last_word_output_sample": 1175040
      },
      {
        "candidate_id": "en-conversation-02-podcast-segment-free-u04-c00",
        "output_offset_sample": 1190400,
        "first_word_output_sample": 1194240,
        "last_word_output_sample": 1455360
      },
      {
        "candidate_id": "en-conversation-02-podcast-segment-free-u05-c00",
        "output_offset_sample": 1470720,
        "first_word_output_sample": 1474560,
        "last_word_output_sample": 2012160
      },
      {
        "candidate_id": "en-conversation-02-podcast-segment-free-u06-c00",
        "output_offset_sample": 2027520,
        "first_word_output_sample": 2031360,
        "last_word_output_sample": 2265600
      },
      {
        "candidate_id": "en-conversation-02-podcast-segment-free-u07-c00",
        "output_offset_sample": 2284800,
        "first_word_output_sample": 2284800,
        "last_word_output_sample": 2507520
      },
      {
        "candidate_id": "en-conversation-02-podcast-segment-free-u08-c00",
        "output_offset_sample": 2522880,
        "first_word_output_sample": 2526720,
        "last_word_output_sample": 2795520
      },
      {
        "candidate_id": "en-conversation-02-podcast-segment-free-u09-c00",
        "output_offset_sample": 2810880,
        "first_word_output_sample": 2814720,
        "last_word_output_sample": 3183360
      },
      {
        "candidate_id": "en-conversation-02-podcast-segment-free-u10-c00",
        "output_offset_sample": 3198720,
        "first_word_output_sample": 3202560,
        "last_word_output_sample": 3459840
      },
      {
        "candidate_id": "en-conversation-02-podcast-segment-free-u11-c00",
        "output_offset_sample": 3475200,
        "first_word_output_sample": 3479040,
        "last_word_output_sample": 3732480
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 77.91,
    "samples": 3739680,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-02-podcast-segment-free-n1.mp3"
  },
  {
    "id": "en-conversation-02-podcast-segment-timed-n1",
    "case": "en-conversation-02",
    "model": "podcast",
    "strategy": "segment",
    "timing": "timed",
    "best_of": 1,
    "block_size": 1,
    "plain": "Imagine a tiny radio studio where the hosts have just discovered this voice model. Please tell me the tiny studio has a very dramatic red recording light. Of course. Humaneness Voice Large supplies the foundation; podcast fine-tuning teaches dialogue. And does the second host interrupt with healthy skepticism, or just politely agree? Healthy skepticism, please. We give each turn its own mood and its own words. I refuse to be dazzled by a cheerful description. Let me hear the voices change. That is why the reference clips sit beside the output. You can compare for yourself. Now that is a demonstration I can enjoy. Keep the little red light on. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/be8bba6062483f0f66e7c553a2ed76cb83aa6b5d719e839557e5d6cdd3c188d2.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-02-podcast-segment-timed-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 218880
      },
      {
        "candidate_id": "en-conversation-02-podcast-segment-timed-u01-c00",
        "output_offset_sample": 238080,
        "first_word_output_sample": 238080,
        "last_word_output_sample": 441600
      },
      {
        "candidate_id": "en-conversation-02-podcast-segment-timed-u02-c00",
        "output_offset_sample": 460800,
        "first_word_output_sample": 460800,
        "last_word_output_sample": 660480
      },
      {
        "candidate_id": "en-conversation-02-podcast-segment-timed-u03-c00",
        "output_offset_sample": 679680,
        "first_word_output_sample": 679680,
        "last_word_output_sample": 883200
      },
      {
        "candidate_id": "en-conversation-02-podcast-segment-timed-u04-c00",
        "output_offset_sample": 898560,
        "first_word_output_sample": 902400,
        "last_word_output_sample": 1117440
      },
      {
        "candidate_id": "en-conversation-02-podcast-segment-timed-u05-c00",
        "output_offset_sample": 1132800,
        "first_word_output_sample": 1136640,
        "last_word_output_sample": 1363200
      },
      {
        "candidate_id": "en-conversation-02-podcast-segment-timed-u06-c00",
        "output_offset_sample": 1382400,
        "first_word_output_sample": 1382400,
        "last_word_output_sample": 1616640
      },
      {
        "candidate_id": "en-conversation-02-podcast-segment-timed-u07-c00",
        "output_offset_sample": 1632000,
        "first_word_output_sample": 1635840,
        "last_word_output_sample": 1850880
      },
      {
        "candidate_id": "en-conversation-02-podcast-segment-timed-u08-c00",
        "output_offset_sample": 1866240,
        "first_word_output_sample": 1870080,
        "last_word_output_sample": 2085120
      },
      {
        "candidate_id": "en-conversation-02-podcast-segment-timed-u09-c00",
        "output_offset_sample": 2100480,
        "first_word_output_sample": 2104320,
        "last_word_output_sample": 2288640
      },
      {
        "candidate_id": "en-conversation-02-podcast-segment-timed-u10-c00",
        "output_offset_sample": 2304000,
        "first_word_output_sample": 2307840,
        "last_word_output_sample": 2526720
      },
      {
        "candidate_id": "en-conversation-02-podcast-segment-timed-u11-c00",
        "output_offset_sample": 2542080,
        "first_word_output_sample": 2545920,
        "last_word_output_sample": 2772480
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 57.91,
    "samples": 2779680,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-02-podcast-segment-timed-n1.mp3"
  },
  {
    "id": "en-conversation-03-base17-block-free-n1",
    "case": "en-conversation-03",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 1,
    "block_size": 4,
    "plain": "Here is the part that interests me: using a short recording to establish each voice. Then keeping those voices distinct while the conversation wanders somewhere unexpected? Yes. The model sees speaker-labelled references and, when available, earlier dialogue. I would love that to work, but long conversations are where voices can drift. We test that directly, turn by turn, rather than trusting the first sentence. Good. I want the delighted host to remain the same person when she becomes doubtful. References guide identity; acting directions guide expression. Neither is a magic guarantee. Somehow the honest explanation is more exciting than the exaggerated one. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/b6eae663a1cda6e3f1c1bc6e161973df1366245dbfd094e4e25191a8c268c9cf.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-03-base17-block-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 890880
      },
      {
        "candidate_id": "en-conversation-03-base17-block-free-u01-c00",
        "output_offset_sample": 906240,
        "first_word_output_sample": 910080,
        "last_word_output_sample": 1739520
      },
      {
        "candidate_id": "en-conversation-03-base17-block-free-u02-c00",
        "output_offset_sample": 1758720,
        "first_word_output_sample": 1758720,
        "last_word_output_sample": 2630400
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 54.95,
    "samples": 2637600,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-03-base17-block-free-n1.mp3"
  },
  {
    "id": "en-conversation-03-base17-block-free-n4",
    "case": "en-conversation-03",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 4,
    "block_size": 4,
    "plain": "Here is the part that interests me: using a short recording to establish each voice. Then keeping those voices distinct while the conversation wanders somewhere unexpected? Yes. The model sees speaker-labelled references and, when available, earlier dialogue. I would love that to work, but long conversations are where voices can drift. We test that directly, turn by turn, rather than trusting the first sentence. Good. I want the delighted host to remain the same person when she becomes doubtful. References guide identity; acting directions guide expression. Neither is a magic guarantee. Somehow the honest explanation is more exciting than the exaggerated one. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/1d03300e2eb551290ee1a12bb80f510df61ff4ac7268e37b67174f561385060c.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-03-base17-block-free-u00-c03",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 913920
      },
      {
        "candidate_id": "en-conversation-03-base17-block-free-u01-c00",
        "output_offset_sample": 929280,
        "first_word_output_sample": 933120,
        "last_word_output_sample": 1762560
      },
      {
        "candidate_id": "en-conversation-03-base17-block-free-u02-c03",
        "output_offset_sample": 1777920,
        "first_word_output_sample": 1781760,
        "last_word_output_sample": 2653440
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 55.43,
    "samples": 2660640,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-03-base17-block-free-n4.mp3"
  },
  {
    "id": "en-conversation-03-base17-block-free-n8",
    "case": "en-conversation-03",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 8,
    "block_size": 4,
    "plain": "Here is the part that interests me: using a short recording to establish each voice. Then keeping those voices distinct while the conversation wanders somewhere unexpected? Yes. The model sees speaker-labelled references and, when available, earlier dialogue. I would love that to work, but long conversations are where voices can drift. We test that directly, turn by turn, rather than trusting the first sentence. Good. I want the delighted host to remain the same person when she becomes doubtful. References guide identity; acting directions guide expression. Neither is a magic guarantee. Somehow the honest explanation is more exciting than the exaggerated one. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/4a60e9afca5c57734fff2f3435e14afb854d709fe6a828e27cb27b2b66d21f01.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-03-base17-block-free-u00-c03",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 913920
      },
      {
        "candidate_id": "en-conversation-03-base17-block-free-u01-c04",
        "output_offset_sample": 933120,
        "first_word_output_sample": 933120,
        "last_word_output_sample": 1814400
      },
      {
        "candidate_id": "en-conversation-03-base17-block-free-u02-c03",
        "output_offset_sample": 1829760,
        "first_word_output_sample": 1833600,
        "last_word_output_sample": 2705280
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 56.51,
    "samples": 2712480,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-03-base17-block-free-n8.mp3"
  },
  {
    "id": "en-conversation-03-base17-block-timed-n1",
    "case": "en-conversation-03",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 1,
    "block_size": 4,
    "plain": "Here is the part that interests me: using a short recording to establish each voice. Then keeping those voices distinct while the conversation wanders somewhere unexpected? Yes. The model sees speaker-labelled references and, when available, earlier dialogue. I would love that to work, but long conversations are where voices can drift. We test that directly, turn by turn, rather than trusting the first sentence. Good. I want the delighted host to remain the same person when she becomes doubtful. References guide identity; acting directions guide expression. Neither is a magic guarantee. Somehow the honest explanation is more exciting than the exaggerated one. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/3836f14c59d6859775cf7233c2b44525c6fd5ac12aa70d66e4684c77cc74e888.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-03-base17-block-timed-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 817920
      },
      {
        "candidate_id": "en-conversation-03-base17-block-timed-u01-c00",
        "output_offset_sample": 833280,
        "first_word_output_sample": 837120,
        "last_word_output_sample": 1635840
      },
      {
        "candidate_id": "en-conversation-03-base17-block-timed-u02-c00",
        "output_offset_sample": 1651200,
        "first_word_output_sample": 1655040,
        "last_word_output_sample": 2526720
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 52.79,
    "samples": 2533920,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-03-base17-block-timed-n1.mp3"
  },
  {
    "id": "en-conversation-03-base17-block-timed-n4",
    "case": "en-conversation-03",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 4,
    "block_size": 4,
    "plain": "Here is the part that interests me: using a short recording to establish each voice. Then keeping those voices distinct while the conversation wanders somewhere unexpected? Yes. The model sees speaker-labelled references and, when available, earlier dialogue. I would love that to work, but long conversations are where voices can drift. We test that directly, turn by turn, rather than trusting the first sentence. Good. I want the delighted host to remain the same person when she becomes doubtful. References guide identity; acting directions guide expression. Neither is a magic guarantee. Somehow the honest explanation is more exciting than the exaggerated one. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/5644f7d0b214c530be4bee292c701c5db9ce9145e5f4fc76ac12b0ed2e6246d8.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-03-base17-block-timed-u00-c02",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 810240
      },
      {
        "candidate_id": "en-conversation-03-base17-block-timed-u01-c01",
        "output_offset_sample": 825600,
        "first_word_output_sample": 829440,
        "last_word_output_sample": 1632000
      },
      {
        "candidate_id": "en-conversation-03-base17-block-timed-u02-c02",
        "output_offset_sample": 1647360,
        "first_word_output_sample": 1651200,
        "last_word_output_sample": 2522880
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 52.71,
    "samples": 2530080,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-03-base17-block-timed-n4.mp3"
  },
  {
    "id": "en-conversation-03-base17-block-timed-n8",
    "case": "en-conversation-03",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 8,
    "block_size": 4,
    "plain": "Here is the part that interests me: using a short recording to establish each voice. Then keeping those voices distinct while the conversation wanders somewhere unexpected? Yes. The model sees speaker-labelled references and, when available, earlier dialogue. I would love that to work, but long conversations are where voices can drift. We test that directly, turn by turn, rather than trusting the first sentence. Good. I want the delighted host to remain the same person when she becomes doubtful. References guide identity; acting directions guide expression. Neither is a magic guarantee. Somehow the honest explanation is more exciting than the exaggerated one. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/6c7f8f1519f10e2393c8c57e1fd359313a0383c3f025a1c440e824ad57e4f598.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-03-base17-block-timed-u00-c06",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 810240
      },
      {
        "candidate_id": "en-conversation-03-base17-block-timed-u01-c04",
        "output_offset_sample": 825600,
        "first_word_output_sample": 829440,
        "last_word_output_sample": 1632000
      },
      {
        "candidate_id": "en-conversation-03-base17-block-timed-u02-c02",
        "output_offset_sample": 1647360,
        "first_word_output_sample": 1651200,
        "last_word_output_sample": 2522880
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 52.71,
    "samples": 2530080,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-03-base17-block-timed-n8.mp3"
  },
  {
    "id": "en-conversation-03-base17-segment-free-n1",
    "case": "en-conversation-03",
    "model": "base17",
    "strategy": "segment",
    "timing": "free",
    "best_of": 1,
    "block_size": 1,
    "plain": "Here is the part that interests me: using a short recording to establish each voice. Then keeping those voices distinct while the conversation wanders somewhere unexpected? Yes. The model sees speaker-labelled references and, when available, earlier dialogue. I would love that to work, but long conversations are where voices can drift. We test that directly, turn by turn, rather than trusting the first sentence. Good. I want the delighted host to remain the same person when she becomes doubtful. References guide identity; acting directions guide expression. Neither is a magic guarantee. Somehow the honest explanation is more exciting than the exaggerated one. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/d3f2b7dc30c07432ac2ede76a0ec904bfa7742f0331b62fb6cc0e65f76c6bce2.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-03-base17-segment-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 288000
      },
      {
        "candidate_id": "en-conversation-03-base17-segment-free-u01-c00",
        "output_offset_sample": 303360,
        "first_word_output_sample": 307200,
        "last_word_output_sample": 591360
      },
      {
        "candidate_id": "en-conversation-03-base17-segment-free-u02-c00",
        "output_offset_sample": 610560,
        "first_word_output_sample": 610560,
        "last_word_output_sample": 902400
      },
      {
        "candidate_id": "en-conversation-03-base17-segment-free-u03-c00",
        "output_offset_sample": 921600,
        "first_word_output_sample": 921600,
        "last_word_output_sample": 1152000
      },
      {
        "candidate_id": "en-conversation-03-base17-segment-free-u04-c00",
        "output_offset_sample": 1171200,
        "first_word_output_sample": 1171200,
        "last_word_output_sample": 1432320
      },
      {
        "candidate_id": "en-conversation-03-base17-segment-free-u05-c00",
        "output_offset_sample": 1447680,
        "first_word_output_sample": 1451520,
        "last_word_output_sample": 1743360
      },
      {
        "candidate_id": "en-conversation-03-base17-segment-free-u06-c00",
        "output_offset_sample": 1758720,
        "first_word_output_sample": 1762560,
        "last_word_output_sample": 2161920
      },
      {
        "candidate_id": "en-conversation-03-base17-segment-free-u07-c00",
        "output_offset_sample": 2181120,
        "first_word_output_sample": 2181120,
        "last_word_output_sample": 2438400
      },
      {
        "candidate_id": "en-conversation-03-base17-segment-free-u08-c00",
        "output_offset_sample": 2453760,
        "first_word_output_sample": 2457600,
        "last_word_output_sample": 2737920
      },
      {
        "candidate_id": "en-conversation-03-base17-segment-free-u09-c00",
        "output_offset_sample": 2753280,
        "first_word_output_sample": 2757120,
        "last_word_output_sample": 3025920
      },
      {
        "candidate_id": "en-conversation-03-base17-segment-free-u10-c00",
        "output_offset_sample": 3045120,
        "first_word_output_sample": 3045120,
        "last_word_output_sample": 3290880
      },
      {
        "candidate_id": "en-conversation-03-base17-segment-free-u11-c00",
        "output_offset_sample": 3306240,
        "first_word_output_sample": 3310080,
        "last_word_output_sample": 3575040
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 74.63,
    "samples": 3582240,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-03-base17-segment-free-n1.mp3"
  },
  {
    "id": "en-conversation-03-base17-segment-timed-n1",
    "case": "en-conversation-03",
    "model": "base17",
    "strategy": "segment",
    "timing": "timed",
    "best_of": 1,
    "block_size": 1,
    "plain": "Here is the part that interests me: using a short recording to establish each voice. Then keeping those voices distinct while the conversation wanders somewhere unexpected? Yes. The model sees speaker-labelled references and, when available, earlier dialogue. I would love that to work, but long conversations are where voices can drift. We test that directly, turn by turn, rather than trusting the first sentence. Good. I want the delighted host to remain the same person when she becomes doubtful. References guide identity; acting directions guide expression. Neither is a magic guarantee. Somehow the honest explanation is more exciting than the exaggerated one. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/2425ad0d0ba09cec937b77be5e640ecc6560d932cfb6629dd3d82978f3bb4541.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-03-base17-segment-timed-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 234240
      },
      {
        "candidate_id": "en-conversation-03-base17-segment-timed-u01-c00",
        "output_offset_sample": 249600,
        "first_word_output_sample": 253440,
        "last_word_output_sample": 422400
      },
      {
        "candidate_id": "en-conversation-03-base17-segment-timed-u02-c00",
        "output_offset_sample": 441600,
        "first_word_output_sample": 441600,
        "last_word_output_sample": 618240
      },
      {
        "candidate_id": "en-conversation-03-base17-segment-timed-u03-c00",
        "output_offset_sample": 633600,
        "first_word_output_sample": 637440,
        "last_word_output_sample": 844800
      },
      {
        "candidate_id": "en-conversation-03-base17-segment-timed-u04-c00",
        "output_offset_sample": 860160,
        "first_word_output_sample": 864000,
        "last_word_output_sample": 1059840
      },
      {
        "candidate_id": "en-conversation-03-base17-segment-timed-u05-c00",
        "output_offset_sample": 1075200,
        "first_word_output_sample": 1079040,
        "last_word_output_sample": 1301760
      },
      {
        "candidate_id": "en-conversation-03-base17-segment-timed-u06-c00",
        "output_offset_sample": 1320960,
        "first_word_output_sample": 1320960,
        "last_word_output_sample": 1516800
      },
      {
        "candidate_id": "en-conversation-03-base17-segment-timed-u07-c00",
        "output_offset_sample": 1536000,
        "first_word_output_sample": 1536000,
        "last_word_output_sample": 1712640
      },
      {
        "candidate_id": "en-conversation-03-base17-segment-timed-u08-c00",
        "output_offset_sample": 1728000,
        "first_word_output_sample": 1731840,
        "last_word_output_sample": 1943040
      },
      {
        "candidate_id": "en-conversation-03-base17-segment-timed-u09-c00",
        "output_offset_sample": 1958400,
        "first_word_output_sample": 1962240,
        "last_word_output_sample": 2142720
      },
      {
        "candidate_id": "en-conversation-03-base17-segment-timed-u10-c00",
        "output_offset_sample": 2161920,
        "first_word_output_sample": 2161920,
        "last_word_output_sample": 2380800
      },
      {
        "candidate_id": "en-conversation-03-base17-segment-timed-u11-c00",
        "output_offset_sample": 2396160,
        "first_word_output_sample": 2400000,
        "last_word_output_sample": 2622720
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 54.79,
    "samples": 2629920,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-03-base17-segment-timed-n1.mp3"
  },
  {
    "id": "en-conversation-03-podcast-block-free-n1",
    "case": "en-conversation-03",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 1,
    "block_size": 4,
    "plain": "Here is the part that interests me: using a short recording to establish each voice. Then keeping those voices distinct while the conversation wanders somewhere unexpected? Yes. The model sees speaker-labelled references and, when available, earlier dialogue. I would love that to work, but long conversations are where voices can drift. We test that directly, turn by turn, rather than trusting the first sentence. Good. I want the delighted host to remain the same person when she becomes doubtful. References guide identity; acting directions guide expression. Neither is a magic guarantee. Somehow the honest explanation is more exciting than the exaggerated one. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/52236832de4d34cbefbea5baa501fe2978ab8ac1207a0a5ec7a2f4bb81a71077.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-03-podcast-block-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 913920
      },
      {
        "candidate_id": "en-conversation-03-podcast-block-free-u01-c00",
        "output_offset_sample": 929280,
        "first_word_output_sample": 933120,
        "last_word_output_sample": 1785600
      },
      {
        "candidate_id": "en-conversation-03-podcast-block-free-u02-c00",
        "output_offset_sample": 1800960,
        "first_word_output_sample": 1804800,
        "last_word_output_sample": 2707200
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 56.55,
    "samples": 2714400,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-03-podcast-block-free-n1.mp3"
  },
  {
    "id": "en-conversation-03-podcast-block-free-n4",
    "case": "en-conversation-03",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 4,
    "block_size": 4,
    "plain": "Here is the part that interests me: using a short recording to establish each voice. Then keeping those voices distinct while the conversation wanders somewhere unexpected? Yes. The model sees speaker-labelled references and, when available, earlier dialogue. I would love that to work, but long conversations are where voices can drift. We test that directly, turn by turn, rather than trusting the first sentence. Good. I want the delighted host to remain the same person when she becomes doubtful. References guide identity; acting directions guide expression. Neither is a magic guarantee. Somehow the honest explanation is more exciting than the exaggerated one. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/d9f594fde18946a24cb486342ff85da45a350c30a43b82c195e6d35a548d9be4.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-03-podcast-block-free-u00-c01",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 898560
      },
      {
        "candidate_id": "en-conversation-03-podcast-block-free-u01-c00",
        "output_offset_sample": 913920,
        "first_word_output_sample": 917760,
        "last_word_output_sample": 1770240
      },
      {
        "candidate_id": "en-conversation-03-podcast-block-free-u02-c02",
        "output_offset_sample": 1785600,
        "first_word_output_sample": 1789440,
        "last_word_output_sample": 2649600
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 55.35,
    "samples": 2656800,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-03-podcast-block-free-n4.mp3"
  },
  {
    "id": "en-conversation-03-podcast-block-free-n8",
    "case": "en-conversation-03",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 8,
    "block_size": 4,
    "plain": "Here is the part that interests me: using a short recording to establish each voice. Then keeping those voices distinct while the conversation wanders somewhere unexpected? Yes. The model sees speaker-labelled references and, when available, earlier dialogue. I would love that to work, but long conversations are where voices can drift. We test that directly, turn by turn, rather than trusting the first sentence. Good. I want the delighted host to remain the same person when she becomes doubtful. References guide identity; acting directions guide expression. Neither is a magic guarantee. Somehow the honest explanation is more exciting than the exaggerated one. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/71881e546c4471f5d8033a0076bf419414784adf08cd6bb038bde7f5709bde7d.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-03-podcast-block-free-u00-c05",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 929280
      },
      {
        "candidate_id": "en-conversation-03-podcast-block-free-u01-c00",
        "output_offset_sample": 944640,
        "first_word_output_sample": 948480,
        "last_word_output_sample": 1800960
      },
      {
        "candidate_id": "en-conversation-03-podcast-block-free-u02-c02",
        "output_offset_sample": 1816320,
        "first_word_output_sample": 1820160,
        "last_word_output_sample": 2680320
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 55.99,
    "samples": 2687520,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-03-podcast-block-free-n8.mp3"
  },
  {
    "id": "en-conversation-03-podcast-block-timed-n1",
    "case": "en-conversation-03",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 1,
    "block_size": 4,
    "plain": "Here is the part that interests me: using a short recording to establish each voice. Then keeping those voices distinct while the conversation wanders somewhere unexpected? Yes. The model sees speaker-labelled references and, when available, earlier dialogue. I would love that to work, but long conversations are where voices can drift. We test that directly, turn by turn, rather than trusting the first sentence. Good. I want the delighted host to remain the same person when she becomes doubtful. References guide identity; acting directions guide expression. Neither is a magic guarantee. Somehow the honest explanation is more exciting than the exaggerated one. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/26f1970fce63c6aefdfc926803b5cafa6113f82898ad3e10d1b035df10abca2a.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-03-podcast-block-timed-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 814080
      },
      {
        "candidate_id": "en-conversation-03-podcast-block-timed-u01-c00",
        "output_offset_sample": 833280,
        "first_word_output_sample": 833280,
        "last_word_output_sample": 1635840
      },
      {
        "candidate_id": "en-conversation-03-podcast-block-timed-u02-c00",
        "output_offset_sample": 1651200,
        "first_word_output_sample": 1655040,
        "last_word_output_sample": 2522880
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 52.71,
    "samples": 2530080,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-03-podcast-block-timed-n1.mp3"
  },
  {
    "id": "en-conversation-03-podcast-block-timed-n4",
    "case": "en-conversation-03",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 4,
    "block_size": 4,
    "plain": "Here is the part that interests me: using a short recording to establish each voice. Then keeping those voices distinct while the conversation wanders somewhere unexpected? Yes. The model sees speaker-labelled references and, when available, earlier dialogue. I would love that to work, but long conversations are where voices can drift. We test that directly, turn by turn, rather than trusting the first sentence. Good. I want the delighted host to remain the same person when she becomes doubtful. References guide identity; acting directions guide expression. Neither is a magic guarantee. Somehow the honest explanation is more exciting than the exaggerated one. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/ba6677c83735ae6f16b4051988e706d8b32253a651657a41e8a698fad5cd337a.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-03-podcast-block-timed-u00-c02",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 810240
      },
      {
        "candidate_id": "en-conversation-03-podcast-block-timed-u01-c02",
        "output_offset_sample": 829440,
        "first_word_output_sample": 829440,
        "last_word_output_sample": 1639680
      },
      {
        "candidate_id": "en-conversation-03-podcast-block-timed-u02-c00",
        "output_offset_sample": 1655040,
        "first_word_output_sample": 1658880,
        "last_word_output_sample": 2526720
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 52.79,
    "samples": 2533920,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-03-podcast-block-timed-n4.mp3"
  },
  {
    "id": "en-conversation-03-podcast-block-timed-n8",
    "case": "en-conversation-03",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 8,
    "block_size": 4,
    "plain": "Here is the part that interests me: using a short recording to establish each voice. Then keeping those voices distinct while the conversation wanders somewhere unexpected? Yes. The model sees speaker-labelled references and, when available, earlier dialogue. I would love that to work, but long conversations are where voices can drift. We test that directly, turn by turn, rather than trusting the first sentence. Good. I want the delighted host to remain the same person when she becomes doubtful. References guide identity; acting directions guide expression. Neither is a magic guarantee. Somehow the honest explanation is more exciting than the exaggerated one. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/5e5d4866f933efe7466f2c632bab5cbf9fcb0256d5e275cb5c4d55bc3d6549e5.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-03-podcast-block-timed-u00-c04",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 810240
      },
      {
        "candidate_id": "en-conversation-03-podcast-block-timed-u01-c04",
        "output_offset_sample": 829440,
        "first_word_output_sample": 829440,
        "last_word_output_sample": 1639680
      },
      {
        "candidate_id": "en-conversation-03-podcast-block-timed-u02-c07",
        "output_offset_sample": 1655040,
        "first_word_output_sample": 1658880,
        "last_word_output_sample": 2526720
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 52.79,
    "samples": 2533920,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-03-podcast-block-timed-n8.mp3"
  },
  {
    "id": "en-conversation-03-podcast-segment-free-n1",
    "case": "en-conversation-03",
    "model": "podcast",
    "strategy": "segment",
    "timing": "free",
    "best_of": 1,
    "block_size": 1,
    "plain": "Here is the part that interests me: using a short recording to establish each voice. Then keeping those voices distinct while the conversation wanders somewhere unexpected? Yes. The model sees speaker-labelled references and, when available, earlier dialogue. I would love that to work, but long conversations are where voices can drift. We test that directly, turn by turn, rather than trusting the first sentence. Good. I want the delighted host to remain the same person when she becomes doubtful. References guide identity; acting directions guide expression. Neither is a magic guarantee. Somehow the honest explanation is more exciting than the exaggerated one. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/3b164ec81a1a16e77deca2681799c8cd0b589741b170ae552ad8a66aa7d1c7e3.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-03-podcast-segment-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 253440
      },
      {
        "candidate_id": "en-conversation-03-podcast-segment-free-u01-c00",
        "output_offset_sample": 268800,
        "first_word_output_sample": 272640,
        "last_word_output_sample": 552960
      },
      {
        "candidate_id": "en-conversation-03-podcast-segment-free-u02-c00",
        "output_offset_sample": 568320,
        "first_word_output_sample": 572160,
        "last_word_output_sample": 833280
      },
      {
        "candidate_id": "en-conversation-03-podcast-segment-free-u03-c00",
        "output_offset_sample": 852480,
        "first_word_output_sample": 852480,
        "last_word_output_sample": 1102080
      },
      {
        "candidate_id": "en-conversation-03-podcast-segment-free-u04-c00",
        "output_offset_sample": 1121280,
        "first_word_output_sample": 1121280,
        "last_word_output_sample": 1359360
      },
      {
        "candidate_id": "en-conversation-03-podcast-segment-free-u05-c00",
        "output_offset_sample": 1374720,
        "first_word_output_sample": 1378560,
        "last_word_output_sample": 1697280
      },
      {
        "candidate_id": "en-conversation-03-podcast-segment-free-u06-c00",
        "output_offset_sample": 1712640,
        "first_word_output_sample": 1716480,
        "last_word_output_sample": 2000640
      },
      {
        "candidate_id": "en-conversation-03-podcast-segment-free-u07-c00",
        "output_offset_sample": 2019840,
        "first_word_output_sample": 2019840,
        "last_word_output_sample": 2257920
      },
      {
        "candidate_id": "en-conversation-03-podcast-segment-free-u08-c00",
        "output_offset_sample": 2273280,
        "first_word_output_sample": 2277120,
        "last_word_output_sample": 2542080
      },
      {
        "candidate_id": "en-conversation-03-podcast-segment-free-u09-c00",
        "output_offset_sample": 2557440,
        "first_word_output_sample": 2561280,
        "last_word_output_sample": 2814720
      },
      {
        "candidate_id": "en-conversation-03-podcast-segment-free-u10-c00",
        "output_offset_sample": 2833920,
        "first_word_output_sample": 2833920,
        "last_word_output_sample": 3110400
      },
      {
        "candidate_id": "en-conversation-03-podcast-segment-free-u11-c00",
        "output_offset_sample": 3125760,
        "first_word_output_sample": 3129600,
        "last_word_output_sample": 3379200
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 70.55,
    "samples": 3386400,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-03-podcast-segment-free-n1.mp3"
  },
  {
    "id": "en-conversation-03-podcast-segment-timed-n1",
    "case": "en-conversation-03",
    "model": "podcast",
    "strategy": "segment",
    "timing": "timed",
    "best_of": 1,
    "block_size": 1,
    "plain": "Here is the part that interests me: using a short recording to establish each voice. Then keeping those voices distinct while the conversation wanders somewhere unexpected? Yes. The model sees speaker-labelled references and, when available, earlier dialogue. I would love that to work, but long conversations are where voices can drift. We test that directly, turn by turn, rather than trusting the first sentence. Good. I want the delighted host to remain the same person when she becomes doubtful. References guide identity; acting directions guide expression. Neither is a magic guarantee. Somehow the honest explanation is more exciting than the exaggerated one. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/83ebc5bdff9dcfc08ac56602096335d4471c23ce3f7b215838c3ab567a356620.mp3",
    "review_candidates": [
      "en-conversation-03-podcast-segment-timed-u02-c00"
    ]
  },
  {
    "id": "en-conversation-04-base17-block-free-n1",
    "case": "en-conversation-04",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 1,
    "block_size": 5,
    "plain": "I can write a requested length next to an individual sentence in the script. You are kidding. Do you mean it lands on an exact millisecond? No. The codec advances in eighty-millisecond frames, and requested timing can still miss. Thank you. I was about to ask it to fit my entire life into three seconds. We compare requested and generated durations, including early endings and frame limits. Then make this sentence unhurried, as if a friend were choosing the right words. That is the idea: control the intention, measure the result, and listen carefully. A reassuringly human approach to a very peculiar little machine. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/fefe3d218b42aa85fd11bd4b95f1f886bd752abbc0ed0296a533453fb5ebd238.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-04-base17-block-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 971520
      },
      {
        "candidate_id": "en-conversation-04-base17-block-free-u01-c00",
        "output_offset_sample": 986880,
        "first_word_output_sample": 990720,
        "last_word_output_sample": 1985280
      },
      {
        "candidate_id": "en-conversation-04-base17-block-free-u02-c00",
        "output_offset_sample": 2000640,
        "first_word_output_sample": 2004480,
        "last_word_output_sample": 2522880
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 52.71,
    "samples": 2530080,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-04-base17-block-free-n1.mp3"
  },
  {
    "id": "en-conversation-04-base17-block-free-n4",
    "case": "en-conversation-04",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 4,
    "block_size": 5,
    "plain": "I can write a requested length next to an individual sentence in the script. You are kidding. Do you mean it lands on an exact millisecond? No. The codec advances in eighty-millisecond frames, and requested timing can still miss. Thank you. I was about to ask it to fit my entire life into three seconds. We compare requested and generated durations, including early endings and frame limits. Then make this sentence unhurried, as if a friend were choosing the right words. That is the idea: control the intention, measure the result, and listen carefully. A reassuringly human approach to a very peculiar little machine. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/485e4df9cdca6107e6311508b4a40a9e524f74ad3b207b9c9cf3ad3a5b2cb409.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-04-base17-block-free-u00-c01",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 998400
      },
      {
        "candidate_id": "en-conversation-04-base17-block-free-u01-c03",
        "output_offset_sample": 1013760,
        "first_word_output_sample": 1017600,
        "last_word_output_sample": 2044800
      },
      {
        "candidate_id": "en-conversation-04-base17-block-free-u02-c01",
        "output_offset_sample": 2060160,
        "first_word_output_sample": 2064000,
        "last_word_output_sample": 2544000
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 53.15,
    "samples": 2551200,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-04-base17-block-free-n4.mp3"
  },
  {
    "id": "en-conversation-04-base17-block-free-n8",
    "case": "en-conversation-04",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 8,
    "block_size": 5,
    "plain": "I can write a requested length next to an individual sentence in the script. You are kidding. Do you mean it lands on an exact millisecond? No. The codec advances in eighty-millisecond frames, and requested timing can still miss. Thank you. I was about to ask it to fit my entire life into three seconds. We compare requested and generated durations, including early endings and frame limits. Then make this sentence unhurried, as if a friend were choosing the right words. That is the idea: control the intention, measure the result, and listen carefully. A reassuringly human approach to a very peculiar little machine. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/59124c7fa69a2957b097329532f11dfb395633a78cfbf7991496d36489607881.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-04-base17-block-free-u00-c04",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 983040
      },
      {
        "candidate_id": "en-conversation-04-base17-block-free-u01-c06",
        "output_offset_sample": 1002240,
        "first_word_output_sample": 1002240,
        "last_word_output_sample": 1885440
      },
      {
        "candidate_id": "en-conversation-04-base17-block-free-u02-c01",
        "output_offset_sample": 1900800,
        "first_word_output_sample": 1904640,
        "last_word_output_sample": 2384640
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 49.83,
    "samples": 2391840,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-04-base17-block-free-n8.mp3"
  },
  {
    "id": "en-conversation-04-base17-block-timed-n1",
    "case": "en-conversation-04",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 1,
    "block_size": 5,
    "plain": "I can write a requested length next to an individual sentence in the script. You are kidding. Do you mean it lands on an exact millisecond? No. The codec advances in eighty-millisecond frames, and requested timing can still miss. Thank you. I was about to ask it to fit my entire life into three seconds. We compare requested and generated durations, including early endings and frame limits. Then make this sentence unhurried, as if a friend were choosing the right words. That is the idea: control the intention, measure the result, and listen carefully. A reassuringly human approach to a very peculiar little machine. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/28472e54a6447e0b97785c31b4eb69422033e3f56459e1497d251353e9eef444.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-04-base17-block-timed-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 1063680
      },
      {
        "candidate_id": "en-conversation-04-base17-block-timed-u01-c00",
        "output_offset_sample": 1079040,
        "first_word_output_sample": 1082880,
        "last_word_output_sample": 2077440
      },
      {
        "candidate_id": "en-conversation-04-base17-block-timed-u02-c00",
        "output_offset_sample": 2092800,
        "first_word_output_sample": 2096640,
        "last_word_output_sample": 2549760
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 53.27,
    "samples": 2556960,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-04-base17-block-timed-n1.mp3"
  },
  {
    "id": "en-conversation-04-base17-block-timed-n4",
    "case": "en-conversation-04",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 4,
    "block_size": 5,
    "plain": "I can write a requested length next to an individual sentence in the script. You are kidding. Do you mean it lands on an exact millisecond? No. The codec advances in eighty-millisecond frames, and requested timing can still miss. Thank you. I was about to ask it to fit my entire life into three seconds. We compare requested and generated durations, including early endings and frame limits. Then make this sentence unhurried, as if a friend were choosing the right words. That is the idea: control the intention, measure the result, and listen carefully. A reassuringly human approach to a very peculiar little machine. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/ab5448413e1cfe2829a8a75c146e5cec4dbc007a1fac6b8dde8e3b9b49ea9f62.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-04-base17-block-timed-u00-c02",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 1067520
      },
      {
        "candidate_id": "en-conversation-04-base17-block-timed-u01-c03",
        "output_offset_sample": 1082880,
        "first_word_output_sample": 1086720,
        "last_word_output_sample": 2081280
      },
      {
        "candidate_id": "en-conversation-04-base17-block-timed-u02-c02",
        "output_offset_sample": 2096640,
        "first_word_output_sample": 2100480,
        "last_word_output_sample": 2549760
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 53.27,
    "samples": 2556960,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-04-base17-block-timed-n4.mp3"
  },
  {
    "id": "en-conversation-04-base17-block-timed-n8",
    "case": "en-conversation-04",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 8,
    "block_size": 5,
    "plain": "I can write a requested length next to an individual sentence in the script. You are kidding. Do you mean it lands on an exact millisecond? No. The codec advances in eighty-millisecond frames, and requested timing can still miss. Thank you. I was about to ask it to fit my entire life into three seconds. We compare requested and generated durations, including early endings and frame limits. Then make this sentence unhurried, as if a friend were choosing the right words. That is the idea: control the intention, measure the result, and listen carefully. A reassuringly human approach to a very peculiar little machine. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/bb4414d17be63a7dcc0e8eb8252c8f3f0325969dc6eca9aecb9bcfaa3c457cf6.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-04-base17-block-timed-u00-c04",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 1067520
      },
      {
        "candidate_id": "en-conversation-04-base17-block-timed-u01-c03",
        "output_offset_sample": 1082880,
        "first_word_output_sample": 1086720,
        "last_word_output_sample": 2081280
      },
      {
        "candidate_id": "en-conversation-04-base17-block-timed-u02-c02",
        "output_offset_sample": 2096640,
        "first_word_output_sample": 2100480,
        "last_word_output_sample": 2549760
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 53.27,
    "samples": 2556960,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-04-base17-block-timed-n8.mp3"
  },
  {
    "id": "en-conversation-04-base17-segment-free-n1",
    "case": "en-conversation-04",
    "model": "base17",
    "strategy": "segment",
    "timing": "free",
    "best_of": 1,
    "block_size": 1,
    "plain": "I can write a requested length next to an individual sentence in the script. You are kidding. Do you mean it lands on an exact millisecond? No. The codec advances in eighty-millisecond frames, and requested timing can still miss. Thank you. I was about to ask it to fit my entire life into three seconds. We compare requested and generated durations, including early endings and frame limits. Then make this sentence unhurried, as if a friend were choosing the right words. That is the idea: control the intention, measure the result, and listen carefully. A reassuringly human approach to a very peculiar little machine. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/5bf5e82ab114d802e6b82c1d94e5bcbe1e6338ac34485520902b4e12c158bcaa.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-04-base17-segment-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 460800
      },
      {
        "candidate_id": "en-conversation-04-base17-segment-free-u01-c00",
        "output_offset_sample": 476160,
        "first_word_output_sample": 480000,
        "last_word_output_sample": 729600
      },
      {
        "candidate_id": "en-conversation-04-base17-segment-free-u02-c00",
        "output_offset_sample": 744960,
        "first_word_output_sample": 748800,
        "last_word_output_sample": 1009920
      },
      {
        "candidate_id": "en-conversation-04-base17-segment-free-u03-c00",
        "output_offset_sample": 1025280,
        "first_word_output_sample": 1029120,
        "last_word_output_sample": 1271040
      },
      {
        "candidate_id": "en-conversation-04-base17-segment-free-u04-c00",
        "output_offset_sample": 1286400,
        "first_word_output_sample": 1290240,
        "last_word_output_sample": 1597440
      },
      {
        "candidate_id": "en-conversation-04-base17-segment-free-u05-c00",
        "output_offset_sample": 1612800,
        "first_word_output_sample": 1616640,
        "last_word_output_sample": 1900800
      },
      {
        "candidate_id": "en-conversation-04-base17-segment-free-u06-c00",
        "output_offset_sample": 1920000,
        "first_word_output_sample": 1920000,
        "last_word_output_sample": 2154240
      },
      {
        "candidate_id": "en-conversation-04-base17-segment-free-u07-c00",
        "output_offset_sample": 2169600,
        "first_word_output_sample": 2173440,
        "last_word_output_sample": 2630400
      },
      {
        "candidate_id": "en-conversation-04-base17-segment-free-u08-c00",
        "output_offset_sample": 2649600,
        "first_word_output_sample": 2649600,
        "last_word_output_sample": 2952960
      },
      {
        "candidate_id": "en-conversation-04-base17-segment-free-u09-c00",
        "output_offset_sample": 2968320,
        "first_word_output_sample": 2972160,
        "last_word_output_sample": 3256320
      },
      {
        "candidate_id": "en-conversation-04-base17-segment-free-u10-c00",
        "output_offset_sample": 3275520,
        "first_word_output_sample": 3275520,
        "last_word_output_sample": 3517440
      },
      {
        "candidate_id": "en-conversation-04-base17-segment-free-u11-c00",
        "output_offset_sample": 3532800,
        "first_word_output_sample": 3536640,
        "last_word_output_sample": 3801600
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 79.35,
    "samples": 3808800,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-04-base17-segment-free-n1.mp3"
  },
  {
    "id": "en-conversation-04-base17-segment-timed-n1",
    "case": "en-conversation-04",
    "model": "base17",
    "strategy": "segment",
    "timing": "timed",
    "best_of": 1,
    "block_size": 1,
    "plain": "I can write a requested length next to an individual sentence in the script. You are kidding. Do you mean it lands on an exact millisecond? No. The codec advances in eighty-millisecond frames, and requested timing can still miss. Thank you. I was about to ask it to fit my entire life into three seconds. We compare requested and generated durations, including early endings and frame limits. Then make this sentence unhurried, as if a friend were choosing the right words. That is the idea: control the intention, measure the result, and listen carefully. A reassuringly human approach to a very peculiar little machine. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/dd1944c120dab2f1fd5b3bcac852c9fb8cedd720f1444399fbb96641e64a94e8.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-04-base17-segment-timed-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 218880
      },
      {
        "candidate_id": "en-conversation-04-base17-segment-timed-u01-c00",
        "output_offset_sample": 234240,
        "first_word_output_sample": 238080,
        "last_word_output_sample": 410880
      },
      {
        "candidate_id": "en-conversation-04-base17-segment-timed-u02-c00",
        "output_offset_sample": 430080,
        "first_word_output_sample": 430080,
        "last_word_output_sample": 637440
      },
      {
        "candidate_id": "en-conversation-04-base17-segment-timed-u03-c00",
        "output_offset_sample": 652800,
        "first_word_output_sample": 656640,
        "last_word_output_sample": 894720
      },
      {
        "candidate_id": "en-conversation-04-base17-segment-timed-u04-c00",
        "output_offset_sample": 913920,
        "first_word_output_sample": 913920,
        "last_word_output_sample": 1105920
      },
      {
        "candidate_id": "en-conversation-04-base17-segment-timed-u05-c00",
        "output_offset_sample": 1121280,
        "first_word_output_sample": 1125120,
        "last_word_output_sample": 1332480
      },
      {
        "candidate_id": "en-conversation-04-base17-segment-timed-u06-c00",
        "output_offset_sample": 1347840,
        "first_word_output_sample": 1351680,
        "last_word_output_sample": 1551360
      },
      {
        "candidate_id": "en-conversation-04-base17-segment-timed-u07-c00",
        "output_offset_sample": 1566720,
        "first_word_output_sample": 1570560,
        "last_word_output_sample": 1724160
      },
      {
        "candidate_id": "en-conversation-04-base17-segment-timed-u08-c00",
        "output_offset_sample": 1739520,
        "first_word_output_sample": 1743360,
        "last_word_output_sample": 1954560
      },
      {
        "candidate_id": "en-conversation-04-base17-segment-timed-u09-c00",
        "output_offset_sample": 1969920,
        "first_word_output_sample": 1973760,
        "last_word_output_sample": 2154240
      },
      {
        "candidate_id": "en-conversation-04-base17-segment-timed-u10-c00",
        "output_offset_sample": 2169600,
        "first_word_output_sample": 2173440,
        "last_word_output_sample": 2384640
      },
      {
        "candidate_id": "en-conversation-04-base17-segment-timed-u11-c00",
        "output_offset_sample": 2400000,
        "first_word_output_sample": 2403840,
        "last_word_output_sample": 2626560
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 54.87,
    "samples": 2633760,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-04-base17-segment-timed-n1.mp3"
  },
  {
    "id": "en-conversation-04-podcast-block-free-n1",
    "case": "en-conversation-04",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 1,
    "block_size": 5,
    "plain": "I can write a requested length next to an individual sentence in the script. You are kidding. Do you mean it lands on an exact millisecond? No. The codec advances in eighty-millisecond frames, and requested timing can still miss. Thank you. I was about to ask it to fit my entire life into three seconds. We compare requested and generated durations, including early endings and frame limits. Then make this sentence unhurried, as if a friend were choosing the right words. That is the idea: control the intention, measure the result, and listen carefully. A reassuringly human approach to a very peculiar little machine. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/67c02cb7da945bd0b34b3c21c952e811456836a1c84d4b2c072c405860ee0304.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-04-podcast-block-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 1094400
      },
      {
        "candidate_id": "en-conversation-04-podcast-block-free-u01-c00",
        "output_offset_sample": 1109760,
        "first_word_output_sample": 1113600,
        "last_word_output_sample": 2200320
      },
      {
        "candidate_id": "en-conversation-04-podcast-block-free-u02-c00",
        "output_offset_sample": 2219520,
        "first_word_output_sample": 2219520,
        "last_word_output_sample": 2711040
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 56.63,
    "samples": 2718240,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-04-podcast-block-free-n1.mp3"
  },
  {
    "id": "en-conversation-04-podcast-block-free-n4",
    "case": "en-conversation-04",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 4,
    "block_size": 5,
    "plain": "I can write a requested length next to an individual sentence in the script. You are kidding. Do you mean it lands on an exact millisecond? No. The codec advances in eighty-millisecond frames, and requested timing can still miss. Thank you. I was about to ask it to fit my entire life into three seconds. We compare requested and generated durations, including early endings and frame limits. Then make this sentence unhurried, as if a friend were choosing the right words. That is the idea: control the intention, measure the result, and listen carefully. A reassuringly human approach to a very peculiar little machine. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/07706b296e0e773d63a81889464207a83c2b34dc9c35b15591fb84f46b65356c.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-04-podcast-block-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 1094400
      },
      {
        "candidate_id": "en-conversation-04-podcast-block-free-u01-c01",
        "output_offset_sample": 1109760,
        "first_word_output_sample": 1113600,
        "last_word_output_sample": 2184960
      },
      {
        "candidate_id": "en-conversation-04-podcast-block-free-u02-c02",
        "output_offset_sample": 2204160,
        "first_word_output_sample": 2204160,
        "last_word_output_sample": 2676480
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 55.91,
    "samples": 2683680,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-04-podcast-block-free-n4.mp3"
  },
  {
    "id": "en-conversation-04-podcast-block-free-n8",
    "case": "en-conversation-04",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 8,
    "block_size": 5,
    "plain": "I can write a requested length next to an individual sentence in the script. You are kidding. Do you mean it lands on an exact millisecond? No. The codec advances in eighty-millisecond frames, and requested timing can still miss. Thank you. I was about to ask it to fit my entire life into three seconds. We compare requested and generated durations, including early endings and frame limits. Then make this sentence unhurried, as if a friend were choosing the right words. That is the idea: control the intention, measure the result, and listen carefully. A reassuringly human approach to a very peculiar little machine. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/f6ff09b81f10531418f48cf638d1259ee13515e06532a71456f883d02683dc50.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-04-podcast-block-free-u00-c06",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 1071360
      },
      {
        "candidate_id": "en-conversation-04-podcast-block-free-u01-c01",
        "output_offset_sample": 1086720,
        "first_word_output_sample": 1090560,
        "last_word_output_sample": 2161920
      },
      {
        "candidate_id": "en-conversation-04-podcast-block-free-u02-c07",
        "output_offset_sample": 2181120,
        "first_word_output_sample": 2181120,
        "last_word_output_sample": 2657280
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 55.51,
    "samples": 2664480,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-04-podcast-block-free-n8.mp3"
  },
  {
    "id": "en-conversation-04-podcast-block-timed-n1",
    "case": "en-conversation-04",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 1,
    "block_size": 5,
    "plain": "I can write a requested length next to an individual sentence in the script. You are kidding. Do you mean it lands on an exact millisecond? No. The codec advances in eighty-millisecond frames, and requested timing can still miss. Thank you. I was about to ask it to fit my entire life into three seconds. We compare requested and generated durations, including early endings and frame limits. Then make this sentence unhurried, as if a friend were choosing the right words. That is the idea: control the intention, measure the result, and listen carefully. A reassuringly human approach to a very peculiar little machine. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/fb2d81648962608ce4844ab1cbfff2daa54d6269b600a0e910bda69575e2f9d8.mp3",
    "review_candidates": [
      "en-conversation-04-podcast-block-timed-u00-c00"
    ]
  },
  {
    "id": "en-conversation-04-podcast-block-timed-n4",
    "case": "en-conversation-04",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 4,
    "block_size": 5,
    "plain": "I can write a requested length next to an individual sentence in the script. You are kidding. Do you mean it lands on an exact millisecond? No. The codec advances in eighty-millisecond frames, and requested timing can still miss. Thank you. I was about to ask it to fit my entire life into three seconds. We compare requested and generated durations, including early endings and frame limits. Then make this sentence unhurried, as if a friend were choosing the right words. That is the idea: control the intention, measure the result, and listen carefully. A reassuringly human approach to a very peculiar little machine. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/d11c0b7a3b3dded4575840c7d4062bff6c9107254893cac5c1c149432936ff55.mp3",
    "review_candidates": [
      "en-conversation-04-podcast-block-timed-u00-c03",
      "en-conversation-04-podcast-block-timed-u01-c03"
    ]
  },
  {
    "id": "en-conversation-04-podcast-block-timed-n8",
    "case": "en-conversation-04",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 8,
    "block_size": 5,
    "plain": "I can write a requested length next to an individual sentence in the script. You are kidding. Do you mean it lands on an exact millisecond? No. The codec advances in eighty-millisecond frames, and requested timing can still miss. Thank you. I was about to ask it to fit my entire life into three seconds. We compare requested and generated durations, including early endings and frame limits. Then make this sentence unhurried, as if a friend were choosing the right words. That is the idea: control the intention, measure the result, and listen carefully. A reassuringly human approach to a very peculiar little machine. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/9a69c4f898d98e875f6c58e47a6bfe4a570630aa13f977ea34ce95bf2b1c7a0d.mp3",
    "review_candidates": [
      "en-conversation-04-podcast-block-timed-u01-c04"
    ]
  },
  {
    "id": "en-conversation-04-podcast-segment-free-n1",
    "case": "en-conversation-04",
    "model": "podcast",
    "strategy": "segment",
    "timing": "free",
    "best_of": 1,
    "block_size": 1,
    "plain": "I can write a requested length next to an individual sentence in the script. You are kidding. Do you mean it lands on an exact millisecond? No. The codec advances in eighty-millisecond frames, and requested timing can still miss. Thank you. I was about to ask it to fit my entire life into three seconds. We compare requested and generated durations, including early endings and frame limits. Then make this sentence unhurried, as if a friend were choosing the right words. That is the idea: control the intention, measure the result, and listen carefully. A reassuringly human approach to a very peculiar little machine. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/d02a791480932f2d5e229aece8928e643148b63e8ed47742295f84fce84171c7.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-04-podcast-segment-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 234240
      },
      {
        "candidate_id": "en-conversation-04-podcast-segment-free-u01-c00",
        "output_offset_sample": 249600,
        "first_word_output_sample": 253440,
        "last_word_output_sample": 460800
      },
      {
        "candidate_id": "en-conversation-04-podcast-segment-free-u02-c00",
        "output_offset_sample": 480000,
        "first_word_output_sample": 480000,
        "last_word_output_sample": 725760
      },
      {
        "candidate_id": "en-conversation-04-podcast-segment-free-u03-c00",
        "output_offset_sample": 741120,
        "first_word_output_sample": 744960,
        "last_word_output_sample": 979200
      },
      {
        "candidate_id": "en-conversation-04-podcast-segment-free-u04-c00",
        "output_offset_sample": 994560,
        "first_word_output_sample": 998400,
        "last_word_output_sample": 1297920
      },
      {
        "candidate_id": "en-conversation-04-podcast-segment-free-u05-c00",
        "output_offset_sample": 1313280,
        "first_word_output_sample": 1317120,
        "last_word_output_sample": 1824000
      },
      {
        "candidate_id": "en-conversation-04-podcast-segment-free-u06-c00",
        "output_offset_sample": 1843200,
        "first_word_output_sample": 1843200,
        "last_word_output_sample": 2108160
      },
      {
        "candidate_id": "en-conversation-04-podcast-segment-free-u07-c00",
        "output_offset_sample": 2123520,
        "first_word_output_sample": 2127360,
        "last_word_output_sample": 2453760
      },
      {
        "candidate_id": "en-conversation-04-podcast-segment-free-u08-c00",
        "output_offset_sample": 2472960,
        "first_word_output_sample": 2472960,
        "last_word_output_sample": 2734080
      },
      {
        "candidate_id": "en-conversation-04-podcast-segment-free-u09-c00",
        "output_offset_sample": 2749440,
        "first_word_output_sample": 2753280,
        "last_word_output_sample": 3002880
      },
      {
        "candidate_id": "en-conversation-04-podcast-segment-free-u10-c00",
        "output_offset_sample": 3018240,
        "first_word_output_sample": 3022080,
        "last_word_output_sample": 3279360
      },
      {
        "candidate_id": "en-conversation-04-podcast-segment-free-u11-c00",
        "output_offset_sample": 3294720,
        "first_word_output_sample": 3298560,
        "last_word_output_sample": 3536640
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 73.83,
    "samples": 3543840,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-04-podcast-segment-free-n1.mp3"
  },
  {
    "id": "en-conversation-04-podcast-segment-timed-n1",
    "case": "en-conversation-04",
    "model": "podcast",
    "strategy": "segment",
    "timing": "timed",
    "best_of": 1,
    "block_size": 1,
    "plain": "I can write a requested length next to an individual sentence in the script. You are kidding. Do you mean it lands on an exact millisecond? No. The codec advances in eighty-millisecond frames, and requested timing can still miss. Thank you. I was about to ask it to fit my entire life into three seconds. We compare requested and generated durations, including early endings and frame limits. Then make this sentence unhurried, as if a friend were choosing the right words. That is the idea: control the intention, measure the result, and listen carefully. A reassuringly human approach to a very peculiar little machine. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/09e6dae0aae1e05ed915d63037cb04b34293720f1fd469224ed2c38e623a8b72.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-04-podcast-segment-timed-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 222720
      },
      {
        "candidate_id": "en-conversation-04-podcast-segment-timed-u01-c00",
        "output_offset_sample": 238080,
        "first_word_output_sample": 241920,
        "last_word_output_sample": 422400
      },
      {
        "candidate_id": "en-conversation-04-podcast-segment-timed-u02-c00",
        "output_offset_sample": 441600,
        "first_word_output_sample": 441600,
        "last_word_output_sample": 652800
      },
      {
        "candidate_id": "en-conversation-04-podcast-segment-timed-u03-c00",
        "output_offset_sample": 668160,
        "first_word_output_sample": 672000,
        "last_word_output_sample": 913920
      },
      {
        "candidate_id": "en-conversation-04-podcast-segment-timed-u04-c00",
        "output_offset_sample": 933120,
        "first_word_output_sample": 933120,
        "last_word_output_sample": 1128960
      },
      {
        "candidate_id": "en-conversation-04-podcast-segment-timed-u05-c00",
        "output_offset_sample": 1144320,
        "first_word_output_sample": 1148160,
        "last_word_output_sample": 1363200
      },
      {
        "candidate_id": "en-conversation-04-podcast-segment-timed-u06-c00",
        "output_offset_sample": 1378560,
        "first_word_output_sample": 1382400,
        "last_word_output_sample": 1578240
      },
      {
        "candidate_id": "en-conversation-04-podcast-segment-timed-u07-c00",
        "output_offset_sample": 1593600,
        "first_word_output_sample": 1597440,
        "last_word_output_sample": 1754880
      },
      {
        "candidate_id": "en-conversation-04-podcast-segment-timed-u08-c00",
        "output_offset_sample": 1770240,
        "first_word_output_sample": 1774080,
        "last_word_output_sample": 1989120
      },
      {
        "candidate_id": "en-conversation-04-podcast-segment-timed-u09-c00",
        "output_offset_sample": 2004480,
        "first_word_output_sample": 2008320,
        "last_word_output_sample": 2192640
      },
      {
        "candidate_id": "en-conversation-04-podcast-segment-timed-u10-c00",
        "output_offset_sample": 2208000,
        "first_word_output_sample": 2211840,
        "last_word_output_sample": 2426880
      },
      {
        "candidate_id": "en-conversation-04-podcast-segment-timed-u11-c00",
        "output_offset_sample": 2442240,
        "first_word_output_sample": 2446080,
        "last_word_output_sample": 2672640
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 55.83,
    "samples": 2679840,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-04-podcast-segment-timed-n1.mp3"
  },
  {
    "id": "en-conversation-05-base17-block-free-n1",
    "case": "en-conversation-05",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 1,
    "block_size": 4,
    "plain": "Picture a listener hearing an English exchange, then German, French, Spanish and Dutch. Same model family, different voices, and a different emotional rhythm in each language? That is what this five-language grid explores, with short and longer conversations. I would start with no references and ask it to invent two hosts. Risky, but fascinating. Then try three, clearly marked as exploratory because training focused on two speakers. I want surprise, warmth, a little teasing, and no one reading the stage directions aloud. Exactly. We publish the full scripts and seeds so the exciting failures remain visible too. All right. I am still skeptical, but now I really want to hear the next turn. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking. Imagine using this for a little science programme, where curiosity matters more than grand claims. And the second host is allowed to say, hold on, explain that again, I have not followed you. Exactly. The exchange should feel responsive, with the same two people returning after every change of mood. Then we have a fair test. Keep the words fixed and let the different generation methods speak for themselves.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/caedda92701cf5d8736fe8181fdecb1f3520ca1cdcf35c9283e7465a301e590a.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-05-base17-block-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 1094400
      },
      {
        "candidate_id": "en-conversation-05-base17-block-free-u01-c00",
        "output_offset_sample": 1109760,
        "first_word_output_sample": 1113600,
        "last_word_output_sample": 1994880
      },
      {
        "candidate_id": "en-conversation-05-base17-block-free-u02-c00",
        "output_offset_sample": 2010240,
        "first_word_output_sample": 2014080,
        "last_word_output_sample": 2989440
      },
      {
        "candidate_id": "en-conversation-05-base17-block-free-u03-c00",
        "output_offset_sample": 3004800,
        "first_word_output_sample": 3008640,
        "last_word_output_sample": 4064640
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 84.83,
    "samples": 4071840,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-05-base17-block-free-n1.mp3"
  },
  {
    "id": "en-conversation-05-base17-block-free-n4",
    "case": "en-conversation-05",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 4,
    "block_size": 4,
    "plain": "Picture a listener hearing an English exchange, then German, French, Spanish and Dutch. Same model family, different voices, and a different emotional rhythm in each language? That is what this five-language grid explores, with short and longer conversations. I would start with no references and ask it to invent two hosts. Risky, but fascinating. Then try three, clearly marked as exploratory because training focused on two speakers. I want surprise, warmth, a little teasing, and no one reading the stage directions aloud. Exactly. We publish the full scripts and seeds so the exciting failures remain visible too. All right. I am still skeptical, but now I really want to hear the next turn. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking. Imagine using this for a little science programme, where curiosity matters more than grand claims. And the second host is allowed to say, hold on, explain that again, I have not followed you. Exactly. The exchange should feel responsive, with the same two people returning after every change of mood. Then we have a fair test. Keep the words fixed and let the different generation methods speak for themselves.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/48887af15acff82f00b698070c19df9d5cfe415e88d3c522bc49a634f1fd75ee.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-05-base17-block-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 1094400
      },
      {
        "candidate_id": "en-conversation-05-base17-block-free-u01-c00",
        "output_offset_sample": 1109760,
        "first_word_output_sample": 1113600,
        "last_word_output_sample": 1994880
      },
      {
        "candidate_id": "en-conversation-05-base17-block-free-u02-c03",
        "output_offset_sample": 2010240,
        "first_word_output_sample": 2014080,
        "last_word_output_sample": 2956800
      },
      {
        "candidate_id": "en-conversation-05-base17-block-free-u03-c02",
        "output_offset_sample": 2972160,
        "first_word_output_sample": 2976000,
        "last_word_output_sample": 3968640
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 82.83,
    "samples": 3975840,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-05-base17-block-free-n4.mp3"
  },
  {
    "id": "en-conversation-05-base17-block-free-n8",
    "case": "en-conversation-05",
    "model": "base17",
    "strategy": "block",
    "timing": "free",
    "best_of": 8,
    "block_size": 4,
    "plain": "Picture a listener hearing an English exchange, then German, French, Spanish and Dutch. Same model family, different voices, and a different emotional rhythm in each language? That is what this five-language grid explores, with short and longer conversations. I would start with no references and ask it to invent two hosts. Risky, but fascinating. Then try three, clearly marked as exploratory because training focused on two speakers. I want surprise, warmth, a little teasing, and no one reading the stage directions aloud. Exactly. We publish the full scripts and seeds so the exciting failures remain visible too. All right. I am still skeptical, but now I really want to hear the next turn. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking. Imagine using this for a little science programme, where curiosity matters more than grand claims. And the second host is allowed to say, hold on, explain that again, I have not followed you. Exactly. The exchange should feel responsive, with the same two people returning after every change of mood. Then we have a fair test. Keep the words fixed and let the different generation methods speak for themselves.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/b5c215ed0732fb3e3d55ace875850741d83c2a80133dd461d13b26b8f66b0540.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-05-base17-block-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 1094400
      },
      {
        "candidate_id": "en-conversation-05-base17-block-free-u01-c07",
        "output_offset_sample": 1113600,
        "first_word_output_sample": 1113600,
        "last_word_output_sample": 1985280
      },
      {
        "candidate_id": "en-conversation-05-base17-block-free-u02-c07",
        "output_offset_sample": 2000640,
        "first_word_output_sample": 2004480,
        "last_word_output_sample": 2858880
      },
      {
        "candidate_id": "en-conversation-05-base17-block-free-u03-c02",
        "output_offset_sample": 2874240,
        "first_word_output_sample": 2878080,
        "last_word_output_sample": 3870720
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 80.79,
    "samples": 3877920,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-05-base17-block-free-n8.mp3"
  },
  {
    "id": "en-conversation-05-base17-block-timed-n1",
    "case": "en-conversation-05",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 1,
    "block_size": 4,
    "plain": "Picture a listener hearing an English exchange, then German, French, Spanish and Dutch. Same model family, different voices, and a different emotional rhythm in each language? That is what this five-language grid explores, with short and longer conversations. I would start with no references and ask it to invent two hosts. Risky, but fascinating. Then try three, clearly marked as exploratory because training focused on two speakers. I want surprise, warmth, a little teasing, and no one reading the stage directions aloud. Exactly. We publish the full scripts and seeds so the exciting failures remain visible too. All right. I am still skeptical, but now I really want to hear the next turn. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking. Imagine using this for a little science programme, where curiosity matters more than grand claims. And the second host is allowed to say, hold on, explain that again, I have not followed you. Exactly. The exchange should feel responsive, with the same two people returning after every change of mood. Then we have a fair test. Keep the words fixed and let the different generation methods speak for themselves.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/0944a0caf4e0330f691723bd9e34d1c7ebac9cff25ac44f36e085ccea39faa19.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-05-base17-block-timed-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 860160
      },
      {
        "candidate_id": "en-conversation-05-base17-block-timed-u01-c00",
        "output_offset_sample": 875520,
        "first_word_output_sample": 879360,
        "last_word_output_sample": 1808640
      },
      {
        "candidate_id": "en-conversation-05-base17-block-timed-u02-c00",
        "output_offset_sample": 1824000,
        "first_word_output_sample": 1827840,
        "last_word_output_sample": 2699520
      },
      {
        "candidate_id": "en-conversation-05-base17-block-timed-u03-c00",
        "output_offset_sample": 2714880,
        "first_word_output_sample": 2718720,
        "last_word_output_sample": 3813120
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 79.59,
    "samples": 3820320,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-05-base17-block-timed-n1.mp3"
  },
  {
    "id": "en-conversation-05-base17-block-timed-n4",
    "case": "en-conversation-05",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 4,
    "block_size": 4,
    "plain": "Picture a listener hearing an English exchange, then German, French, Spanish and Dutch. Same model family, different voices, and a different emotional rhythm in each language? That is what this five-language grid explores, with short and longer conversations. I would start with no references and ask it to invent two hosts. Risky, but fascinating. Then try three, clearly marked as exploratory because training focused on two speakers. I want surprise, warmth, a little teasing, and no one reading the stage directions aloud. Exactly. We publish the full scripts and seeds so the exciting failures remain visible too. All right. I am still skeptical, but now I really want to hear the next turn. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking. Imagine using this for a little science programme, where curiosity matters more than grand claims. And the second host is allowed to say, hold on, explain that again, I have not followed you. Exactly. The exchange should feel responsive, with the same two people returning after every change of mood. Then we have a fair test. Keep the words fixed and let the different generation methods speak for themselves.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/2f6ed3eb0ca594aafa938b80651f7f1beec5c7cf04a5cff794500b52fb87ccfa.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-05-base17-block-timed-u00-c01",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 860160
      },
      {
        "candidate_id": "en-conversation-05-base17-block-timed-u01-c03",
        "output_offset_sample": 875520,
        "first_word_output_sample": 879360,
        "last_word_output_sample": 1808640
      },
      {
        "candidate_id": "en-conversation-05-base17-block-timed-u02-c02",
        "output_offset_sample": 1824000,
        "first_word_output_sample": 1827840,
        "last_word_output_sample": 2703360
      },
      {
        "candidate_id": "en-conversation-05-base17-block-timed-u03-c02",
        "output_offset_sample": 2718720,
        "first_word_output_sample": 2722560,
        "last_word_output_sample": 3816960
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 79.67,
    "samples": 3824160,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-05-base17-block-timed-n4.mp3"
  },
  {
    "id": "en-conversation-05-base17-block-timed-n8",
    "case": "en-conversation-05",
    "model": "base17",
    "strategy": "block",
    "timing": "timed",
    "best_of": 8,
    "block_size": 4,
    "plain": "Picture a listener hearing an English exchange, then German, French, Spanish and Dutch. Same model family, different voices, and a different emotional rhythm in each language? That is what this five-language grid explores, with short and longer conversations. I would start with no references and ask it to invent two hosts. Risky, but fascinating. Then try three, clearly marked as exploratory because training focused on two speakers. I want surprise, warmth, a little teasing, and no one reading the stage directions aloud. Exactly. We publish the full scripts and seeds so the exciting failures remain visible too. All right. I am still skeptical, but now I really want to hear the next turn. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking. Imagine using this for a little science programme, where curiosity matters more than grand claims. And the second host is allowed to say, hold on, explain that again, I have not followed you. Exactly. The exchange should feel responsive, with the same two people returning after every change of mood. Then we have a fair test. Keep the words fixed and let the different generation methods speak for themselves.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/45dc8fbc3a9a5ba98c117457a5971da11fb28eff30a5e072059e727f821006d5.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-05-base17-block-timed-u00-c06",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 860160
      },
      {
        "candidate_id": "en-conversation-05-base17-block-timed-u01-c07",
        "output_offset_sample": 875520,
        "first_word_output_sample": 879360,
        "last_word_output_sample": 1808640
      },
      {
        "candidate_id": "en-conversation-05-base17-block-timed-u02-c04",
        "output_offset_sample": 1824000,
        "first_word_output_sample": 1827840,
        "last_word_output_sample": 2691840
      },
      {
        "candidate_id": "en-conversation-05-base17-block-timed-u03-c02",
        "output_offset_sample": 2707200,
        "first_word_output_sample": 2711040,
        "last_word_output_sample": 3805440
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 79.43,
    "samples": 3812640,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-05-base17-block-timed-n8.mp3"
  },
  {
    "id": "en-conversation-05-base17-segment-free-n1",
    "case": "en-conversation-05",
    "model": "base17",
    "strategy": "segment",
    "timing": "free",
    "best_of": 1,
    "block_size": 1,
    "plain": "Picture a listener hearing an English exchange, then German, French, Spanish and Dutch. Same model family, different voices, and a different emotional rhythm in each language? That is what this five-language grid explores, with short and longer conversations. I would start with no references and ask it to invent two hosts. Risky, but fascinating. Then try three, clearly marked as exploratory because training focused on two speakers. I want surprise, warmth, a little teasing, and no one reading the stage directions aloud. Exactly. We publish the full scripts and seeds so the exciting failures remain visible too. All right. I am still skeptical, but now I really want to hear the next turn. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking. Imagine using this for a little science programme, where curiosity matters more than grand claims. And the second host is allowed to say, hold on, explain that again, I have not followed you. Exactly. The exchange should feel responsive, with the same two people returning after every change of mood. Then we have a fair test. Keep the words fixed and let the different generation methods speak for themselves.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/c5c0d51e4eb1b9506b1419d2ddcec1a47b274095009a61ed8f475c0f95758746.mp3",
    "review_candidates": [
      "en-conversation-05-base17-segment-free-u07-c00",
      "en-conversation-05-base17-segment-free-u11-c00"
    ]
  },
  {
    "id": "en-conversation-05-base17-segment-timed-n1",
    "case": "en-conversation-05",
    "model": "base17",
    "strategy": "segment",
    "timing": "timed",
    "best_of": 1,
    "block_size": 1,
    "plain": "Picture a listener hearing an English exchange, then German, French, Spanish and Dutch. Same model family, different voices, and a different emotional rhythm in each language? That is what this five-language grid explores, with short and longer conversations. I would start with no references and ask it to invent two hosts. Risky, but fascinating. Then try three, clearly marked as exploratory because training focused on two speakers. I want surprise, warmth, a little teasing, and no one reading the stage directions aloud. Exactly. We publish the full scripts and seeds so the exciting failures remain visible too. All right. I am still skeptical, but now I really want to hear the next turn. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking. Imagine using this for a little science programme, where curiosity matters more than grand claims. And the second host is allowed to say, hold on, explain that again, I have not followed you. Exactly. The exchange should feel responsive, with the same two people returning after every change of mood. Then we have a fair test. Keep the words fixed and let the different generation methods speak for themselves.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/f886d80e85b1ff00ff15bd51839aea3efc55030c2050305ee980449ffa877027.mp3",
    "review_candidates": [
      "en-conversation-05-base17-segment-timed-u06-c00",
      "en-conversation-05-base17-segment-timed-u07-c00"
    ]
  },
  {
    "id": "en-conversation-05-podcast-block-free-n1",
    "case": "en-conversation-05",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 1,
    "block_size": 4,
    "plain": "Picture a listener hearing an English exchange, then German, French, Spanish and Dutch. Same model family, different voices, and a different emotional rhythm in each language? That is what this five-language grid explores, with short and longer conversations. I would start with no references and ask it to invent two hosts. Risky, but fascinating. Then try three, clearly marked as exploratory because training focused on two speakers. I want surprise, warmth, a little teasing, and no one reading the stage directions aloud. Exactly. We publish the full scripts and seeds so the exciting failures remain visible too. All right. I am still skeptical, but now I really want to hear the next turn. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking. Imagine using this for a little science programme, where curiosity matters more than grand claims. And the second host is allowed to say, hold on, explain that again, I have not followed you. Exactly. The exchange should feel responsive, with the same two people returning after every change of mood. Then we have a fair test. Keep the words fixed and let the different generation methods speak for themselves.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/e5e92d3122a9ea5f37d51cad8c47a85806af2c0f5e7b72d30a70868ef18e55d6.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-05-podcast-block-free-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 971520
      },
      {
        "candidate_id": "en-conversation-05-podcast-block-free-u01-c00",
        "output_offset_sample": 990720,
        "first_word_output_sample": 990720,
        "last_word_output_sample": 1900800
      },
      {
        "candidate_id": "en-conversation-05-podcast-block-free-u02-c00",
        "output_offset_sample": 1916160,
        "first_word_output_sample": 1920000,
        "last_word_output_sample": 2853120
      },
      {
        "candidate_id": "en-conversation-05-podcast-block-free-u03-c00",
        "output_offset_sample": 2868480,
        "first_word_output_sample": 2872320,
        "last_word_output_sample": 3901440
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 81.43,
    "samples": 3908640,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-05-podcast-block-free-n1.mp3"
  },
  {
    "id": "en-conversation-05-podcast-block-free-n4",
    "case": "en-conversation-05",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 4,
    "block_size": 4,
    "plain": "Picture a listener hearing an English exchange, then German, French, Spanish and Dutch. Same model family, different voices, and a different emotional rhythm in each language? That is what this five-language grid explores, with short and longer conversations. I would start with no references and ask it to invent two hosts. Risky, but fascinating. Then try three, clearly marked as exploratory because training focused on two speakers. I want surprise, warmth, a little teasing, and no one reading the stage directions aloud. Exactly. We publish the full scripts and seeds so the exciting failures remain visible too. All right. I am still skeptical, but now I really want to hear the next turn. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking. Imagine using this for a little science programme, where curiosity matters more than grand claims. And the second host is allowed to say, hold on, explain that again, I have not followed you. Exactly. The exchange should feel responsive, with the same two people returning after every change of mood. Then we have a fair test. Keep the words fixed and let the different generation methods speak for themselves.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/be0f092da1cdab86741ae0ace12a2ca9dd2ba32aa7a9b0847179204e65b5603e.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-05-podcast-block-free-u00-c02",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 967680
      },
      {
        "candidate_id": "en-conversation-05-podcast-block-free-u01-c02",
        "output_offset_sample": 986880,
        "first_word_output_sample": 986880,
        "last_word_output_sample": 1950720
      },
      {
        "candidate_id": "en-conversation-05-podcast-block-free-u02-c00",
        "output_offset_sample": 1966080,
        "first_word_output_sample": 1969920,
        "last_word_output_sample": 2903040
      },
      {
        "candidate_id": "en-conversation-05-podcast-block-free-u03-c00",
        "output_offset_sample": 2918400,
        "first_word_output_sample": 2922240,
        "last_word_output_sample": 3951360
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 82.47,
    "samples": 3958560,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-05-podcast-block-free-n4.mp3"
  },
  {
    "id": "en-conversation-05-podcast-block-free-n8",
    "case": "en-conversation-05",
    "model": "podcast",
    "strategy": "block",
    "timing": "free",
    "best_of": 8,
    "block_size": 4,
    "plain": "Picture a listener hearing an English exchange, then German, French, Spanish and Dutch. Same model family, different voices, and a different emotional rhythm in each language? That is what this five-language grid explores, with short and longer conversations. I would start with no references and ask it to invent two hosts. Risky, but fascinating. Then try three, clearly marked as exploratory because training focused on two speakers. I want surprise, warmth, a little teasing, and no one reading the stage directions aloud. Exactly. We publish the full scripts and seeds so the exciting failures remain visible too. All right. I am still skeptical, but now I really want to hear the next turn. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking. Imagine using this for a little science programme, where curiosity matters more than grand claims. And the second host is allowed to say, hold on, explain that again, I have not followed you. Exactly. The exchange should feel responsive, with the same two people returning after every change of mood. Then we have a fair test. Keep the words fixed and let the different generation methods speak for themselves.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/96f27983cce59ed1ac9a64e3c62744405e45f7bb7277dc6f8f9a5dffb54b193d.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-05-podcast-block-free-u00-c02",
        "output_offset_sample": 0,
        "first_word_output_sample": 0,
        "last_word_output_sample": 967680
      },
      {
        "candidate_id": "en-conversation-05-podcast-block-free-u01-c05",
        "output_offset_sample": 983040,
        "first_word_output_sample": 986880,
        "last_word_output_sample": 1946880
      },
      {
        "candidate_id": "en-conversation-05-podcast-block-free-u02-c00",
        "output_offset_sample": 1962240,
        "first_word_output_sample": 1966080,
        "last_word_output_sample": 2899200
      },
      {
        "candidate_id": "en-conversation-05-podcast-block-free-u03-c05",
        "output_offset_sample": 2914560,
        "first_word_output_sample": 2918400,
        "last_word_output_sample": 3924480
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 81.91,
    "samples": 3931680,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-05-podcast-block-free-n8.mp3"
  },
  {
    "id": "en-conversation-05-podcast-block-timed-n1",
    "case": "en-conversation-05",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 1,
    "block_size": 4,
    "plain": "Picture a listener hearing an English exchange, then German, French, Spanish and Dutch. Same model family, different voices, and a different emotional rhythm in each language? That is what this five-language grid explores, with short and longer conversations. I would start with no references and ask it to invent two hosts. Risky, but fascinating. Then try three, clearly marked as exploratory because training focused on two speakers. I want surprise, warmth, a little teasing, and no one reading the stage directions aloud. Exactly. We publish the full scripts and seeds so the exciting failures remain visible too. All right. I am still skeptical, but now I really want to hear the next turn. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking. Imagine using this for a little science programme, where curiosity matters more than grand claims. And the second host is allowed to say, hold on, explain that again, I have not followed you. Exactly. The exchange should feel responsive, with the same two people returning after every change of mood. Then we have a fair test. Keep the words fixed and let the different generation methods speak for themselves.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/fbcd4c8a1c1ab699be549cc6d9e21d7f6cdf153a1c810c0c887f440c831fabb7.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-05-podcast-block-timed-u00-c00",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 860160
      },
      {
        "candidate_id": "en-conversation-05-podcast-block-timed-u01-c00",
        "output_offset_sample": 879360,
        "first_word_output_sample": 879360,
        "last_word_output_sample": 1820160
      },
      {
        "candidate_id": "en-conversation-05-podcast-block-timed-u02-c00",
        "output_offset_sample": 1835520,
        "first_word_output_sample": 1839360,
        "last_word_output_sample": 2703360
      },
      {
        "candidate_id": "en-conversation-05-podcast-block-timed-u03-c00",
        "output_offset_sample": 2722560,
        "first_word_output_sample": 2722560,
        "last_word_output_sample": 3820800
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 79.75,
    "samples": 3828000,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-05-podcast-block-timed-n1.mp3"
  },
  {
    "id": "en-conversation-05-podcast-block-timed-n4",
    "case": "en-conversation-05",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 4,
    "block_size": 4,
    "plain": "Picture a listener hearing an English exchange, then German, French, Spanish and Dutch. Same model family, different voices, and a different emotional rhythm in each language? That is what this five-language grid explores, with short and longer conversations. I would start with no references and ask it to invent two hosts. Risky, but fascinating. Then try three, clearly marked as exploratory because training focused on two speakers. I want surprise, warmth, a little teasing, and no one reading the stage directions aloud. Exactly. We publish the full scripts and seeds so the exciting failures remain visible too. All right. I am still skeptical, but now I really want to hear the next turn. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking. Imagine using this for a little science programme, where curiosity matters more than grand claims. And the second host is allowed to say, hold on, explain that again, I have not followed you. Exactly. The exchange should feel responsive, with the same two people returning after every change of mood. Then we have a fair test. Keep the words fixed and let the different generation methods speak for themselves.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/4d8cd46cadde385ee1fa8f945f398223323b83148a9e5954ab7778ad46856651.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-05-podcast-block-timed-u00-c01",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 860160
      },
      {
        "candidate_id": "en-conversation-05-podcast-block-timed-u01-c02",
        "output_offset_sample": 875520,
        "first_word_output_sample": 879360,
        "last_word_output_sample": 1812480
      },
      {
        "candidate_id": "en-conversation-05-podcast-block-timed-u02-c03",
        "output_offset_sample": 1831680,
        "first_word_output_sample": 1831680,
        "last_word_output_sample": 2711040
      },
      {
        "candidate_id": "en-conversation-05-podcast-block-timed-u03-c00",
        "output_offset_sample": 2730240,
        "first_word_output_sample": 2730240,
        "last_word_output_sample": 3828480
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 79.91,
    "samples": 3835680,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-05-podcast-block-timed-n4.mp3"
  },
  {
    "id": "en-conversation-05-podcast-block-timed-n8",
    "case": "en-conversation-05",
    "model": "podcast",
    "strategy": "block",
    "timing": "timed",
    "best_of": 8,
    "block_size": 4,
    "plain": "Picture a listener hearing an English exchange, then German, French, Spanish and Dutch. Same model family, different voices, and a different emotional rhythm in each language? That is what this five-language grid explores, with short and longer conversations. I would start with no references and ask it to invent two hosts. Risky, but fascinating. Then try three, clearly marked as exploratory because training focused on two speakers. I want surprise, warmth, a little teasing, and no one reading the stage directions aloud. Exactly. We publish the full scripts and seeds so the exciting failures remain visible too. All right. I am still skeptical, but now I really want to hear the next turn. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking. Imagine using this for a little science programme, where curiosity matters more than grand claims. And the second host is allowed to say, hold on, explain that again, I have not followed you. Exactly. The exchange should feel responsive, with the same two people returning after every change of mood. Then we have a fair test. Keep the words fixed and let the different generation methods speak for themselves.",
    "ready": true,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/9a6d7be3320cae3f7cd744f486dfb4263f73b4ea050ef4018e5fb86107092f92.mp3",
    "review_candidates": [],
    "unit_maps": [
      {
        "candidate_id": "en-conversation-05-podcast-block-timed-u00-c01",
        "output_offset_sample": 0,
        "first_word_output_sample": 3840,
        "last_word_output_sample": 860160
      },
      {
        "candidate_id": "en-conversation-05-podcast-block-timed-u01-c02",
        "output_offset_sample": 875520,
        "first_word_output_sample": 879360,
        "last_word_output_sample": 1812480
      },
      {
        "candidate_id": "en-conversation-05-podcast-block-timed-u02-c06",
        "output_offset_sample": 1827840,
        "first_word_output_sample": 1831680,
        "last_word_output_sample": 2699520
      },
      {
        "candidate_id": "en-conversation-05-podcast-block-timed-u03-c00",
        "output_offset_sample": 2718720,
        "first_word_output_sample": 2718720,
        "last_word_output_sample": 3816960
      }
    ],
    "word_to_word_gap_ms": 400,
    "duration_s": 79.67,
    "samples": 3824160,
    "all_inserted_gaps_sample_exact": true,
    "audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/editing_v1/montages/en-conversation-05-podcast-block-timed-n8.mp3"
  },
  {
    "id": "en-conversation-05-podcast-segment-free-n1",
    "case": "en-conversation-05",
    "model": "podcast",
    "strategy": "segment",
    "timing": "free",
    "best_of": 1,
    "block_size": 1,
    "plain": "Picture a listener hearing an English exchange, then German, French, Spanish and Dutch. Same model family, different voices, and a different emotional rhythm in each language? That is what this five-language grid explores, with short and longer conversations. I would start with no references and ask it to invent two hosts. Risky, but fascinating. Then try three, clearly marked as exploratory because training focused on two speakers. I want surprise, warmth, a little teasing, and no one reading the stage directions aloud. Exactly. We publish the full scripts and seeds so the exciting failures remain visible too. All right. I am still skeptical, but now I really want to hear the next turn. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking. Imagine using this for a little science programme, where curiosity matters more than grand claims. And the second host is allowed to say, hold on, explain that again, I have not followed you. Exactly. The exchange should feel responsive, with the same two people returning after every change of mood. Then we have a fair test. Keep the words fixed and let the different generation methods speak for themselves.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/452c7cd565ddd0047df0450d809d6304738aae4a361f080207d9e1bc9ad1334d.mp3",
    "review_candidates": [
      "en-conversation-05-podcast-segment-free-u07-c00"
    ]
  },
  {
    "id": "en-conversation-05-podcast-segment-timed-n1",
    "case": "en-conversation-05",
    "model": "podcast",
    "strategy": "segment",
    "timing": "timed",
    "best_of": 1,
    "block_size": 1,
    "plain": "Picture a listener hearing an English exchange, then German, French, Spanish and Dutch. Same model family, different voices, and a different emotional rhythm in each language? That is what this five-language grid explores, with short and longer conversations. I would start with no references and ask it to invent two hosts. Risky, but fascinating. Then try three, clearly marked as exploratory because training focused on two speakers. I want surprise, warmth, a little teasing, and no one reading the stage directions aloud. Exactly. We publish the full scripts and seeds so the exciting failures remain visible too. All right. I am still skeptical, but now I really want to hear the next turn. There is something lovely about a conversation that leaves room for a real reaction. Yes, a tiny hesitation can say more than another perfectly polished sentence. So let us compare the voices carefully, including the moments where the illusion breaks. Agreed. I want warmth and character, and I still want to recognize who is talking. Imagine using this for a little science programme, where curiosity matters more than grand claims. And the second host is allowed to say, hold on, explain that again, I have not followed you. Exactly. The exchange should feel responsive, with the same two people returning after every change of mood. Then we have a fair test. Keep the words fixed and let the different generation methods speak for themselves.",
    "ready": false,
    "source_selection_unchanged": true,
    "metrics_scope": "Edited montage not rescored; do not copy source quality/identity/WER scores.",
    "previous_audio": "https://huggingface.co/datasets/laion/Humaneness-Podcast-Demo-Audio/resolve/main/segment_block_pilot_20261107/montages/92fd080a949c5361c62f04ed8de5c6f035b40e2074021082b04a1da55f199e94.mp3",
    "review_candidates": [
      "en-conversation-05-podcast-segment-timed-u06-c00",
      "en-conversation-05-podcast-segment-timed-u07-c00"
    ]
  }
]