Multilingual Automatic Speech Recognition on CommonVoice (test)
9.86WERProposed (MoE + modified CIF)
Evaluation Results
| Method | Links | |
|---|---|---|
| Proposed (MoE + modified CIF)Training data duration (h)=80002026.06 | 9.86 | |
| Whisper-large-v32026.06 | 12.53 | |
| Proposed (MoE + modified CIF)Training data duration (h)=15002026.06 | 13.87 | |
| LLM-ASR + MoE ProjectorTraining data duration (h)=15002026.06 | 14.48 | |
| LLM-ASR + CIF DownsamplerTraining data duration (h)=15002026.06 | 18.45 | |
| LLM-ASR BaselineTraining data duration (h)=15002026.06 | 19.57 |