List your company

Speaker-Attributed Automatic Speech Recognition on AISHELL-4 (test)

0.2004cpCERMOSS Transcribe Diarize
-2.20958414.05780830.325246.592592Jan 4, 2026Jan 22, 2026Feb 9, 2026Feb 28, 2026Mar 18, 2026Apr 5, 2026Apr 24, 2026

Evaluation results

MethodLinks
MOSS Transcribe Diarize2026.010.20040.15430.0461———
DoubaoModel=Speech Recognition Model2026.010.27860.18180.0968———
ElevenLabsModel=Scribe v12026.010.37950.19580.1836———
Gemini 2.5 Pro2026.010.53420.4270.1072———
Speaker-Reasoner Multi-turn w/ SACStage=32026.0414.7313.830.95.26——
Speaker-Reasoner Multi-turnStage=22026.0414.9313.831.15.19——
Speaker-Reasoner BaseStage=12026.0416.5414.042.56.24——
DM-ASR (S2SND) (CN+EN 2900h)Parms=1.7B, Training Data / Hours=CN+EN 2900h, Diarization Source=S2SND, Time Prediction Source=LLM-predicted, Label Perturbation Status=None2026.0417.66——10.564.1618.1
SpeakerLMArchitecture=End-to-End Baselines, Training Data=7638.95h2026.0418.3717.171.2———
SpeakerLM (7639h)Parms=7B, Training Data / Hours=7639h2026.0418.37—————
DM-ASR (DiariZen) (CN+EN 2900h)Parms=0.6B, Training Data / Hours=CN+EN 2900h, Diarization Source=DiariZen, Time Prediction Source=LLM-predicted, Label Perturbation Status=None2026.0418.96——12.01—19.67
DM-ASR (DiariZen) (CN+EN 2900h)Parms=1.7B, Training Data / Hours=CN+EN 2900h, Diarization Source=DiariZen, Time Prediction Source=LLM-predicted, Label Perturbation Status=None2026.0419.01——12.01—19.73
DM-ASR (S2SND) (CN 1300h)Parms=1.7B, Training Data / Hours=CN 1300h, Diarization Source=S2SND, Time Prediction Source=LLM-predicted, Label Perturbation Status=None2026.0419.21——10.56—19.94
Qwen3-Omni + SOT sftArchitecture=End-to-End, Stage=Ablation baseline2026.0419.5917.651.94———
DM-ASR (DiariZen) (CN 1300h)Parms=0.6B, Training Data / Hours=CN 1300h, Diarization Source=DiariZen, Time Prediction Source=LLM-predicted, Label Perturbation Status=None2026.0419.69——12—20.4
DM-ASR (DiariZen) (CN 1300h)Parms=1.7B, Training Data / Hours=CN 1300h, Diarization Source=DiariZen, Time Prediction Source=LLM-predicted, Label Perturbation Status=None2026.0419.85——12.01—20.79
MOSS Transcribe DiarizeParms=-2026.0420.04—————
DM-ASR (DiariZen) (CN 630h)Parms=0.6B, Training Data / Hours=CN 630h, Diarization Source=DiariZen, Time Prediction Source=LLM-predicted, Label Perturbation Status=None2026.0421.58——12.01—23.04
Qwen3-Omni-30B-A3B-InstructArchitecture=End-to-End Baselines2026.0422.2214.467.7632.42——
Speaker-Reasoner Multi-turn 7BScale=7B2026.0422.9115.317.69.38——
SpeakerLM (2270h)Parms=7B, Training Data / Hours=2270h2026.0423.1—————
VibeVoice-ASR(>9400h)Parms=7B, Training Data / Hours=>9400h2026.0424.99——6.77—25.35
Gemini-2.5-ProArchitecture=End-to-End Baselines, Access=Closed-source model2026.0425.1119.815.336.07——
SpeakerLM (694h)Parms=7B, Training Data / Hours=694h2026.0425.28—————
Speaker-Reasoner Base 7BScale=7B2026.0425.615.659.9512——
SpeakerLMArchitecture=End-to-End Baselines, Training Data=212.25h2026.0426.1417.758.39———
Pyannote3.1 + ParaformerArchitecture=Cascade Baselines2026.0426.2419.187.068.1——
VibeVoice-ASRArchitecture=End-to-End Baselines2026.0426.322.3410.88——
Gemini-3.0-ProParms=-2026.0427.43——22.03—54.17
Gemini-2.5-ProParms=-2026.0431.59——15.32—35.96
DiariZen+Whisper-large-v3Parms=1.5B2026.0434.42——11.97—37.4
TagSpeech-AlimeetingArchitecture=End-to-End Baselines2026.0453.4435.717.7437.51——
Qwen2.5-Omni-7BArchitecture=End-to-End Baselines2026.0460.4533.3727.0885.68——