Spoken-to-formal ASR on WenetSpeech Formal (test)
0.1606CERFormalASR-1.7B
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| FormalASR-1.7BModel Scale=1.7B, Training Mode=Supervised Fine-tuning (SFT)2026.05 | 0.1606 | 0.8896 | 0.9439 | |
| FormalASR-0.6BModel Scale=0.6B, Training Mode=Supervised Fine-tuning (SFT)2026.05 | 0.177 | 0.8769 | 0.9359 | |
| Qwen3-ASR-1.7BModel Scale=1.7B, Training Mode=Verbatim2026.05 | 0.246 | 0.8571 | 0.9268 | |
| Qwen3-ASR-0.6BModel Scale=0.6B, Training Mode=Verbatim2026.05 | 0.2581 | 0.8463 | 0.9198 | |
| Whisper large-v3Training Mode=Verbatim2026.05 | 0.3631 | 0.7393 | 0.8538 |