Automatic Speech Recognition on ASRU 2019 (test)
2.88Match Error Rate (MER)Interactive ASR
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Interactive ASRInteraction Loop=102026.04 | 2.88 | 20.88 | 0.82 | |
| Interactive ASRInteraction Loop=32026.04 | 3.09 | 22.08 | 3.06 | |
| Interactive ASRInteraction Loop=22026.04 | 3.21 | 23.04 | 4.59 | |
| Interactive ASRInteraction Loop=12026.04 | 3.59 | 25.22 | 8.1 | |
| Whisper-large-v3 + ft full model#Train. Params.=1541.6 M, pre-training (pt)=true, Tuning=full-parameter fine-tuning2025.09 | 6.6 | — | — | |
| Interactive ASRInteraction Loop=02026.04 | 6.6 | 38.85 | 26.89 | |
| Whisper-medium + ft full model#Train. Params.=762.3 M, pre-training (pt)=true, Tuning=full-parameter fine-tuning2025.09 | 6.8 | — | — | |
| Whisper-small + ft full model + w/ FA-LID-UB#Train. Params.=240.6 M, pre-training (pt)=true, Tuning=full-parameter fine-tuning2025.09 | 8.7 | — | — | |
| Whisper-small + ft full model + w/ LAL#Train. Params.=240.6 M, pre-training (pt)=true, Tuning=full-parameter fine-tuning2025.09 | 8.7 | — | — | |
| Whisper-small + ft full model#Train. Params.=240.6 M, pre-training (pt)=true, Tuning=full-parameter fine-tuning2025.09 | 8.8 | — | — | |
| Whisper-small + ft full model + w/ FA-LID-LB#Train. Params.=240.6 M, pre-training (pt)=true, Tuning=full-parameter fine-tuning2025.09 | 8.8 | — | — | |
| Whisper-small + ft full model + w/ CTC-LID#Train. Params.=240.6 M, pre-training (pt)=true, Tuning=full-parameter fine-tuning2025.09 | 8.9 | — | — | |
| Whisper-small + ft encoder-only#Train. Params.=87.0 M, pre-training (pt)=true, Tuning=full-parameter fine-tuning2025.09 | 9.8 | — | — | |
| Whisper-small + adapt. w/ Bi-LoRA#Train. Params.=4.4 M, pre-training (pt)=true, Tuning=adaptation tuning2025.09 | 10 | — | — | |
| Whisper-small + adapt. w/ LoRA#Train. Params.=4.4 M, pre-training (pt)=true, Tuning=adaptation tuning2025.09 | 10.2 | — | — | |
| Whisper-small + ft decoder-only#Train. Params.=153.6 M, pre-training (pt)=true, Tuning=full-parameter fine-tuning2025.09 | 10.3 | — | — | |
| Whisper-large-v3pre-training (pt)=true2025.09 | 10.3 | — | — | |
| Conformer + LLM GER#Train. Params.=48.3 M, pre-training (pt)=false2025.09 | 11 | — | — | |
| E-Branchformer + CAMEL#Train. Params.=55.3 M, pre-training (pt)=false2025.09 | 11.4 | — | — | |
| ConExtBiMamba*#Train. Params.=54.6 M, pre-training (pt)=false2025.09 | 11.5 | — | — | |
| Conformer + LAL#Train. Params.=48.3 M, pre-training (pt)=false2025.09 | 11.7 | — | — | |
| Conformer + XLS-R-0.3B#Train. Params.=48.3 M, pre-training (pt)=true2025.09 | 11.8 | — | — | |
| Conformer + LPB#Train. Params.=79.9 M, pre-training (pt)=false2025.09 | 11.8 | — | — | |
| Conformer + ILB#Train. Params.=79.9 M, pre-training (pt)=false2025.09 | 11.8 | — | — | |
| E-Branchformer#Train. Params.=39.9 M, pre-training (pt)=false2025.09 | 11.8 | — | — | |
| Branchformer#Train. Params.=39.0 M, pre-training (pt)=false2025.09 | 11.9 | — | — | |
| Conformer + WavLM-large#Train. Params.=48.3 M, pre-training (pt)=true2025.09 | 12 | — | — | |
| Whisper-mediumpre-training (pt)=true2025.09 | 12.1 | — | — | |
| Conformer#Train. Params.=48.3 M, pre-training (pt)=false2025.09 | 12.8 | — | — | |
| Whisper-small + ft full model + w/ SPT*#Train. Params.=240.7 M, pre-training (pt)=true, Tuning=full-parameter fine-tuning2025.09 | 12.8 | — | — | |
| Transformer#Train. Params.=29.8 M, pre-training (pt)=false2025.09 | 13.1 | — | — | |
| Whisper-smallpre-training (pt)=true2025.09 | 24.9 | — | — |