manhpv26/khmer-yt-voice-dataset

---
language:
- km
license: cc-by-sa-4.0
task_categories:
- automatic-speech-recognition
- audio-classification
size_categories:
- 100K<n<1M
---

# Khmer YouTube Voice Dataset

Full audio + diarization metadata from Khmer YouTube videos.

| Stat | Value |
|------|-------|
| Videos | 3,945 |
| Total turns | 16,021 |
| Total duration | 1033.4 hours |
| Format | WAV (original sample rate) |

## Structure

```
data/
shard_0000.tar.gz
shard_0001.tar.gz
...
manifest.jsonl
```

Mỗi shard chứa ~20 videos, mỗi video gồm:
```
video_id/
full.wav # audio nguyên gốc
metadata.json # video info + turns (speaker, start, end, transcript)
```

## manifest.jsonl

| Field | Description |
|-------|-------------|
| `video_id` | YouTube video ID |
| `shard` | Shard chứa video |
| `duration_sec` | Thời lượng video (seconds) |
| `num_turns` | Số turns (speaker segments) |
| `title` | Video title |
| `channel` | YouTube channel |
| `source_url` | YouTube URL |