Automatic Speech Recognition on MLC-SLM (dev)
10.99WER/CERWhisper
Evaluation Results
| Method | Links | |
|---|---|---|
| Whispertuning=Full-fine-tuned2026.01 | 10.99 | |
| Whispertuning=LoRA-fine-tuned2026.01 | 11.4 | |
| NTU-Speechlab2026.01 | 11.57 | |
| Proposed Speech-LLM2026.01 | 11.74 | |
| Seewoo2026.01 | 12.73 | |
| SHNU-mASR2026.01 | 13.39 | |
| Proposed (MoE + modified CIF)Training data duration (h)=15002026.06 | 15.27 | |
| Proposed (MoE + modified CIF)Training data duration (h)=80002026.06 | 15.45 | |
| LLM-ASR + MoE ProjectorTraining data duration (h)=15002026.06 | 16.1 | |
| LLM-ASR + CIF DownsamplerTraining data duration (h)=15002026.06 | 18.95 | |
| Whisper-large-v32026.06 | 21.48 | |
| mHuBERT-CTC (baseline)Params(M)=97, TrainingData=MLC-500h, language-known=false, inference_type=single2026.01 | 22.5 | |
| LLM-ASR BaselineTraining data duration (h)=15002026.06 | 23.26 | |
| mHuBERT-CTC-LIDLoRAParams(M)=107, TrainingData=MLC-500h, language-known=true, inference_type=single2026.01 | 24.6 | |
| mHuBERT-CTC-HLoRAParams(M)=102, TrainingData=MLC-500h, language-known=true, inference_type=single2026.01 | 26 | |
| mHuBERT-CTC-HLoRAParams(M)=102, TrainingData=MLC-500h, language-known=false, inference_type=single2026.01 | 26.3 | |
| mHuBERT-CTC-LIDLoRAParams(M)=107, TrainingData=MLC-500h, language-known=false, inference_type=double2026.01 | 26.6 | |
| mHuBERTmode=CTC2026.01 | 29.67 | |
| Whisper-smallParams(M)=244, TrainingData=680,000h, language-known=false, inference_type=single2026.01 | 29.7 | |
| Whisper-baseParams(M)=74, TrainingData=680,000h, language-known=false, inference_type=single2026.01 | 37.2 | |
| mHuBERT-CTC (baseline)Params(M)=97, TrainingData=MSR-1500h, language-known=false, inference_type=single2026.01 | 44.5 |