Multi-speaker Automatic Speech Recognition on AMI SDM (test)
22.07tcpWERDM-ASR (S2SND) (CN+EN 2900h)
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| DM-ASR (S2SND) (CN+EN 2900h)Parms=1.7B, Label perturbation=None, Diarization source=diarization-provided speakers, Time prediction=LLM-predicted times2026.04 | 22.07 | 13.72 | 21.26 | 5.95 | — | — | — | |
| DM-ASR (DiariZen) (CN+EN 2900h)Parms=1.7B, Label perturbation=None, Diarization source=diarization-provided speakers, Time prediction=LLM-predicted times2026.04 | 24.97 | 14.83 | 23.45 | — | — | — | — | |
| DM-ASR (DiariZen) (CN+EN 2900h)Parms=0.6B, Label perturbation=None, Diarization source=diarization-provided speakers, Time prediction=LLM-predicted times2026.04 | 25.1 | 14.83 | 23.47 | — | — | — | — | |
| Topline (B2)Label=Tight, Tightening strategy=N/A2026.06 | 25.93 | — | — | — | 11.83 | 7.59 | 6.51 | |
| Proposed (P2)Label=Loose, Tightening strategy=VAD2026.06 | 26.31 | — | — | — | 11.88 | 7.66 | 6.77 | |
| DM-ASR (DiariZen) (EN 1600h)Parms=0.6B, Label perturbation=None, Diarization source=diarization-provided speakers, Time prediction=LLM-predicted times2026.04 | 26.6 | 14.83 | 25.05 | — | — | — | — | |
| Proposed (P3)Label=Loose, Tightening strategy=SC2026.06 | 26.68 | — | — | — | 12.77 | 7.55 | 6.36 | |
| Proposed (P1)Label=Loose, Tightening strategy=Basic2026.06 | 28.21 | — | — | — | 13.51 | 8.47 | 6.23 | |
| Baseline (B1)Label=Loose, Tightening strategy=N/A2026.06 | 28.3 | — | — | — | 12.7 | 9.05 | 6.55 | |
| DM-ASR (DiariZen) (EN 630h)Parms=0.6B, Label perturbation=None, Diarization source=diarization-provided speakers, Time prediction=LLM-predicted times2026.04 | 28.9 | 14.83 | 26.89 | — | — | — | — | |
| VibeVoice-ASRParms=7B2026.04 | 29.8 | 13.43 | 28.82 | — | — | — | — | |
| Gemini-2.5-ProParms=-2026.04 | 41.39 | 23.79 | 34.78 | — | — | — | — | |
| DiariZen + Whisper-large-v3Parms=1.5B2026.04 | 49.06 | 14.61 | 43.91 | — | — | — | — | |
| Gemini-3.0-ProParms=-2026.04 | 64.86 | 43.04 | 26.91 | — | — | — | — | |
| Pyannote + Whisper-large-v3Parms=1.5B2026.04 | — | 23.05 | 43.57 | — | — | — | — |