yt_health_audio_5_22s (segments)

This dataset contains aligned audio segments (FLAC 16 kHz mono) and minimally refined transcripts for clinical/medical Portuguese speech.

Storage modes

• files: audio segments stored individually under audio/<video_id>/<id>.flac (previewable in the Hub UI).
• shards: audio segments packed into TARs under shards/shard-*.tar with matching manifests under metadata/part-*.jsonl. This is efficient for large datasets and aligns with WebDataset-style consumption.

JSONL schema (both modes)

{
  "id": "E9qrcBzn1s8_3119500_3133640",
  "split": "train",
  "text": "...",
  "language": "pt",
  "start_ms": 3119500,
  "end_ms": 3133640,
  "segment_duration_s": 14.14,
  "source_duration_s": 14.14,
  "avg_conf": 0.90,
  "low_conf_frac": 0.23,
  "video_id": "E9qrcBzn1s8",
  "channel_id": "UC...",
  "video_title": "...",
  "channel_title": "...",
  "youtube_url": "https://www.youtube.com/watch?v=...",
  "audio_shard": "shards/shard-000000.tar",   // shards mode
  "audio_key": "audio/E9qrcBzn1s8/E9qrcBzn1s8_3119500_3133640.flac",
  "hf_audio_path": "hf://datasets/igorcouto/yt_health_audio_5_22s/audio/E9qrcBzn1s8/E9qrcBzn1s8_3119500_3133640.flac"
}

In shards mode, hf_audio_path is a hint of the form hf://datasets/igorcouto/yt_health_audio_5_22s/shards/shard-000000.tar#audio/....