dia-MsdWild-test / README.md
ggfox00000's picture
Upload MSDWild test: few=490, many=177
56e78e1 verified
metadata
license: cc-by-nc-4.0
task_categories:
  - automatic-speech-recognition
  - voice-activity-detection
language:
  - en
  - zh
  - multilingual
size_categories:
  - n<1K
pretty_name: MSDWild (val = test)  Speaker Diarization in the Wild
tags:
  - speaker-diarization
  - diarization
  - msdwild
  - in-the-wild
  - multimodal
  - rttm
  - audio
annotations_creators:
  - expert-generated
source_datasets:
  - extended|msdwild
dataset_info:
  - config_name: few
    features:
      - name: audio
        dtype: audio
      - name: rttm
        dtype: string
      - name: file_id
        dtype: string
    splits:
      - name: test
        num_examples: 490
  - config_name: many
    features:
      - name: audio
        dtype: audio
      - name: rttm
        dtype: string
      - name: file_id
        dtype: string
    splits:
      - name: test
        num_examples: 177

MSDWild — validation / test split (speaker diarization in the wild)

Copie du split d'évaluation de MSDWild (Liu et al., Interspeech 2022), un corpus de diarisation in the wild multimodal construit à partir de vidéos réelles (talk-shows, interviews, débats…). MSDWild fournit deux sets de validation utilisés comme test de facto par la communauté :

  • few : 2 à 4 locuteurs par session (490 fichiers) — tâche "classique"
  • many : 5+ locuteurs par session (177 fichiers) — tâche conversationnelle dense, beaucoup plus difficile

Les RTTMs sont ceux distribués officiellement sur le repo upstream. Les audios sont les WAV officiels (Google Drive, 7.56 GB).

Contenu

Sous-split Fichiers Profil
few 490 2–4 locuteurs, segments courts
many 177 5+ locuteurs, densité de tour de parole élevée
  • Audio : WAV 16 kHz mono (tels que distribués par MSDWild upstream)
  • Annotations : RTTM standard pyannote
  • Langues : multilingue (principalement anglais et chinois, sources web)
  • Licence : CC-BY-NC-4.0 — recherche uniquement (voir l'accord de licence MSDWild upstream)

Structure

dia-MsdWild-test/
├── README.md
├── audio/
│   ├── few/<file_id>.wav     # 490
│   └── many/<file_id>.wav    # 177
└── rttm/
    ├── few/<file_id>.rttm    # 490
    └── many/<file_id>.rttm   # 177

Chaque ligne RTTM suit le format standard :

SPEAKER <file_id> 1 <start_sec> <duration_sec> <NA> <NA> <speaker_id> <NA> <NA>

Utilisation

Téléchargement

from huggingface_hub import snapshot_download
root = snapshot_download("ggfox00000/dia-MsdWild-test", repo_type="dataset")
# root/audio/few/*.wav, root/rttm/few/*.rttm, root/audio/many/*.wav, ...

Évaluation DER (pyannote)

from pathlib import Path
from pyannote.audio import Pipeline
from pyannote.metrics.diarization import DiarizationErrorRate
from pyannote.database.util import load_rttm

pipe = Pipeline.from_pretrained("pyannote/speaker-diarization-3.1")

for subset in ("few", "many"):
    metric = DiarizationErrorRate()
    audio_dir = Path(root) / "audio" / subset
    rttm_dir  = Path(root) / "rttm"  / subset
    for wav in sorted(audio_dir.glob("*.wav")):
        hyp = pipe(str(wav))
        ref = next(iter(load_rttm(str(rttm_dir / f"{wav.stem}.rttm")).values()))
        metric(ref, hyp)
    print(f"{subset}: DER = {abs(metric):.3f}")

Source

Licence

CC-BY-NC-4.0 — usage non commercial, recherche uniquement. Se référer à l'accord de licence MSDWild upstream (MSDWILD_license_agreement.pdf) avant toute réutilisation.

Citation

@inproceedings{liu22t_interspeech,
  author    = {Tao Liu and Shuai Fan and Xu Xiang and Hongbo Song and Shaoxiong Lin and Jiaqi Sun and Tianyuan Han and Siyuan Chen and Binwei Yao and Sen Liu and Yifei Wu and Yanmin Qian and Kai Yu},
  title     = {{MSDWild: Multi-modal Speaker Diarization Dataset in the Wild}},
  booktitle = {Interspeech},
  year      = {2022},
  pages     = {1476--1480},
  doi       = {10.21437/Interspeech.2022-10466},
}