sea-asr-id-data release (id-v4 labels)

.gitattributes

@@ -0,0 +1,60 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.avro filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.lz4 filter=lfs diff=lfs merge=lfs -text
+*.mds filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+# Audio files - uncompressed
+*.pcm filter=lfs diff=lfs merge=lfs -text
+*.sam filter=lfs diff=lfs merge=lfs -text
+*.raw filter=lfs diff=lfs merge=lfs -text
+# Audio files - compressed
+*.aac filter=lfs diff=lfs merge=lfs -text
+*.flac filter=lfs diff=lfs merge=lfs -text
+*.mp3 filter=lfs diff=lfs merge=lfs -text
+*.ogg filter=lfs diff=lfs merge=lfs -text
+*.wav filter=lfs diff=lfs merge=lfs -text
+# Image files - uncompressed
+*.bmp filter=lfs diff=lfs merge=lfs -text
+*.gif filter=lfs diff=lfs merge=lfs -text
+*.png filter=lfs diff=lfs merge=lfs -text
+*.tiff filter=lfs diff=lfs merge=lfs -text
+# Image files - compressed
+*.jpg filter=lfs diff=lfs merge=lfs -text
+*.jpeg filter=lfs diff=lfs merge=lfs -text
+*.webp filter=lfs diff=lfs merge=lfs -text
+# Video files - compressed
+*.mp4 filter=lfs diff=lfs merge=lfs -text
+*.webm filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,68 @@
+---
+license: apache-2.0
+language: [id]
+task_categories: [automatic-speech-recognition]
+pretty_name: sea-asr-id-data
+size_categories: [1M<n<10M]
+---
+
+# sea-asr-id-data — Bahasa Indonesia speech transcripts
+
+Transcripts and source ids. No audio.
+
+2,287.4 h of machine-labelled Bahasa Indonesia speech from 35,198 YouTube videos in YODAS: 1,971,625 segments, each
+with the video id, start and end time, and a transcript that two ASR models agree on. These are the YODAS training
+labels of [PixelML/sea-asr-id](https://huggingface.co/PixelML/sea-asr-id) (id-v4):
+
+| `label_round` | Segments | Hours | What |
+|---|---|---|---|
+| `v3` | 1,367,189 | 1,901.2 | segments from YODAS id000 + id100 |
+| `v4-short-clips` | 604,436 | 386.3 | 1–3 s clips from the same YODAS subsets, same rule |
+
+The model also trained on FLEURS and Common Voice; those rows are not in this dataset (get them from their owners;
+Common Voice terms do not allow re-hosting). One video is not included because it also appears in a public test set
+(GigaSpeech 2 Indonesian test).
+
+## Label quality: machine labels
+
+Each segment was transcribed by Qwen3-ASR-1.7B and by Whisper large-v3. We keep the Qwen3-ASR text only where the
+two agree: character error rate between them ≤ 0.2 (`agreement_cer`). No person checked them. The YODAS caption
+text is not included.
+
+## Fields
+
+| Field | Meaning |
+|---|---|
+| `source`, `source_id` | `youtube`, YouTube video id |
+| `segment_id` | YODAS utterance id `<video_id>-<seg>-<start_cs>-<end_cs>` |
+| `start`, `end` | seconds in the source video (from the YODAS utterance id) |
+| `text_norm` | transcript, lower case, no punctuation, numbers as words |
+| `text_raw` | Qwen3-ASR output as stored by the labelling pipeline |
+| `label_source` | labelling models and the agreement rule |
+| `agreement_cer` | CER between the Qwen3-ASR and Whisper transcripts |
+| `licence` | CC-BY by the creator, via YODAS (CC-BY-3.0) |
+| `yodas_subset` | `yodas-id000` or `yodas-id100` |
+| `label_round` | `v3` or `v4-short-clips` (see above) |
+
+`ATTRIBUTION.tsv` / `.jsonl` lists every source video id. YODAS publishes no channel names, so each row gives the video id, its URL and the YODAS source; `channel` and `channel_id` are empty in this dataset.
+
+## Get the audio
+
+`python fetch_audio.py --data data/ --out audio/` downloads each source video with yt-dlp and cuts the segments.
+YODAS also ships the audio: https://huggingface.co/datasets/espnet/yodas (match on `segment_id`).
+
+## Credits
+
+- Audio: YODAS (ESPnet/CMU, 52c5a1b9), subsets id000 and id100 — https://huggingface.co/datasets/espnet/yodas
+- Labels: Qwen3-ASR-1.7B (Alibaba Qwen, 7278e1e7); Whisper large-v3 (OpenAI) via faster-whisper (SYSTRAN)
+- Audio-rebuild pattern: PleIAs, YouTube-Commons — https://huggingface.co/datasets/PleIAs/YouTube-Commons
+- Recipe: Quan Nguyen (@qnguyen3), QUAN-Transcribe — https://github.com/qnguyen3/QUAN-Transcribe (4f45e02)
+
+## Licence
+
+Labels: Apache-2.0. Audio: not included; each video stays under its creator's CC-BY licence (see `ATTRIBUTION`).
+
+## Limits
+
+- Machine labels; the agreement filter favours clear speech. No speaker, gender or dialect balance was measured.
+- Lower case, no punctuation.

data/data-00000.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8ff763666a34400ffeebbe49e5d976e527fc3b9d1d769d44276020ddfa537a6f
+size 11669077

data/data-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d1ac5f396fb910f341e4df411d1538c185cf2aae88d2c1c6aa779ca9feeef5d6
+size 11712543

data/data-00002.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e877fa2df3485af7182744795c71fe59dcff37c8aa107a70f778109ec1aecb5f
+size 11731916

data/data-00003.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:595cec4612bf61f3867f2db0a1cc3fc2201e14b4ffb6e3ec066e0c6aa309c466
+size 11781449

data/data-00004.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:514ccb2b190df928197e609f5ea60ec5d2b93f5c1201eda6b8dac62ea89751f1
+size 11852460

data/data-00005.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:79a7afc86d0ece089319b975590937d5e1dc8d63275e3de6341735305df0472d
+size 11540888

data/data-00006.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b69454b2739277d77bce4927e768785f1b7d516d9ddefae91aab4c11e3f8276
+size 11687257

data/data-00007.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1112963bd5c5cf90a7fd1f7fb8c58104ba2790dde4cc9140d14fd6a0cfb60938
+size 11754108

data/data-00008.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2d0a325ed12bee785620246bbd12782cd45d2b9e8e7c658c36539f5b313ed9ff
+size 11692231

data/data-00009.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aa5d652db347be7dc8cb673ceb8b87b676c9228839a9bef9265f8437b304795f
+size 10054734

fetch_audio.py

@@ -0,0 +1,77 @@
+"""Rebuild audio for the sea-asr label datasets from the source videos. The datasets ship no audio.
+
+Pattern from PleIAs, YouTube-Commons (https://huggingface.co/datasets/PleIAs/YouTube-Commons): publish transcripts
+and source ids; users fetch the audio themselves from the original Creative Commons videos.
+
+    pip install pyarrow soundfile numpy yt-dlp       # and ffmpeg on PATH
+    python fetch_audio.py --data data/ --out audio/ [--limit-videos 10]
+
+Writes <out>/<segment_id>.flac (16 kHz mono) and <out>/manifest.jsonl ({segment_id, audio, text, start, end}).
+Each video stays under its creator's CC-BY licence: keep the attribution (ATTRIBUTION.tsv) with any copy you make.
+Videos can be deleted or changed upstream; segments whose source is gone are skipped and counted.
+"""
+
+import argparse
+import glob
+import json
+import os
+import subprocess
+import tempfile
+
+import numpy as np
+import pyarrow.parquet as pq
+import soundfile as sf
+
+SR = 16000
+
+
+def fetch_video(video_id: str, tmp: str) -> np.ndarray | None:
+    """Best audio of one YouTube video as 16 kHz mono float32, or None if unavailable."""
+    path = os.path.join(tmp, video_id)
+    r = subprocess.run(["yt-dlp", "-q", "--no-warnings", "-f", "bestaudio", "-o", path + ".%(ext)s",
+                        f"https://www.youtube.com/watch?v={video_id}"], capture_output=True)
+    files = glob.glob(path + ".*")
+    if r.returncode != 0 or not files:
+        return None
+    raw = subprocess.run(["ffmpeg", "-nostdin", "-loglevel", "error", "-i", files[0], "-ac", "1", "-ar", str(SR),
+                          "-f", "f32le", "-"], capture_output=True).stdout
+    for f in files:
+        os.remove(f)
+    return np.frombuffer(raw, dtype=np.float32) if raw else None
+
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--data", required=True, help="folder with the dataset's data-*.parquet files")
+    ap.add_argument("--out", required=True)
+    ap.add_argument("--limit-videos", type=int, default=0)
+    a = ap.parse_args()
+    os.makedirs(a.out, exist_ok=True)
+    cols = ["source_id", "segment_id", "start", "end", "text_norm"]
+    rows = [r for f in sorted(glob.glob(os.path.join(a.data, "*.parquet"))) for r in pq.read_table(f, columns=cols).to_pylist()]
+    by_vid: dict[str, list[dict]] = {}
+    for r in rows:
+        by_vid.setdefault(r["source_id"], []).append(r)
+    vids = sorted(by_vid)[: a.limit_videos or None]
+    done = gone = 0
+    with tempfile.TemporaryDirectory() as tmp, open(os.path.join(a.out, "manifest.jsonl"), "a", encoding="utf-8") as man:
+        for vid in vids:
+            wav = fetch_video(vid, tmp)
+            if wav is None:
+                gone += len(by_vid[vid])
+                continue
+            for r in by_vid[vid]:
+                seg = wav[int(r["start"] * SR): int(r["end"] * SR)]
+                if len(seg) < SR // 2:
+                    gone += 1
+                    continue
+                path = os.path.join(a.out, r["segment_id"] + ".flac")
+                sf.write(path, seg, SR)
+                man.write(json.dumps({"segment_id": r["segment_id"], "audio": path, "text": r["text_norm"],
+                                      "start": r["start"], "end": r["end"]}, ensure_ascii=False) + "\n")
+                done += 1
+    print(f"segments written {done}, skipped (video gone or too short) {gone}")
+
+
+if __name__ == "__main__":
+    main()

stats.json

@@ -0,0 +1,43 @@
+{
+ "rows": 1971625,
+ "videos": 35198,
+ "hours": 2287.4,
+ "by_round": {
+  "v3": {
+   "rows": 1367189,
+   "hours": 1901.2
+  },
+  "v4-short-clips": {
+   "rows": 604436,
+   "hours": 386.3
+  }
+ },
+ "files": 10,
+ "columns": [
+  "source",
+  "source_id",
+  "segment_id",
+  "start",
+  "end",
+  "text_norm",
+  "text_raw",
+  "label_source",
+  "agreement_cer",
+  "licence",
+  "yodas_subset",
+  "label_round"
+ ],
+ "checks": {
+  "banned_columns": [],
+  "duplicate_segment_ids": 0,
+  "excluded_yt_present": 0
+ },
+ "input": {
+  "rows": 1972168,
+  "kept": 1971625,
+  "bad_id": 0,
+  "not_yodas": 0,
+  "excluded_yt": 543,
+  "hours": 2287.4
+ }
+}