ggfox00000/dia-voxconverse-test

VoxConverse — test split (speaker diarization)

Copie du split test de VoxConverse v0.3 mise en forme pour les benchmarks de diarisation (pyannote, NeMo, etc.). 232 fichiers audio + 232 RTTM de référence.

Contenu

• 232 enregistrements (TV/YouTube anglais, multi-speakers, réunions & débats)
• Audio : WAV 16 kHz, mono
• Annotations : RTTM (Rich Transcription Time Marked)
• Langue : anglais (en)
• Licence : CC-BY-4.0 (identique à VoxConverse upstream)

Structure

dia-voxconverse-test/
├── audio/test/<file_id>.wav       # 232 fichiers
└── rttm/test/<file_id>.rttm       # 232 fichiers

Chaque ligne RTTM :

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

Utilisation

Téléchargement direct

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

Évaluation pyannote (DER)

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")
metric = DiarizationErrorRate()

audio_dir = Path(root) / "audio/test"
rttm_dir  = Path(root) / "rttm/test"
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"DER: {abs(metric):.3f}")

Source

• VoxConverse v0.3 — Chung et al. 2020 https://github.com/joonson/voxconverse

Citation

@inproceedings{chung20voxconverse,
  title = {{VoxConverse: a Free-flowing Speech Dataset for Speaker Diarisation}},
  author = {Chung, Joon Son and Huh, Jaesung and Nagrani, Arsha and Afouras, Triantafyllos and Zisserman, Andrew},
  booktitle = {Interspeech},
  year = {2020},
}

Licence

CC-BY-4.0 (identique à la source upstream).