Automatic Speech Recognition on VOiCES
1.81RM1Step-Audio-2-mini
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Step-Audio-2-miniModel type=Open-source2026.05 | 1.81 | 1.98 | 23.25 | 15.19 | 10.56 | |
| Kimi-AudioModel type=Open-source2026.05 | 2.1 | 2.23 | 26.95 | 15.13 | 11.6 | |
| Mega-ASRModel type=Our model2026.05 | 2.36 | 2.43 | 15.13 | 9.46 | 7.35 | |
| Mega-ASR w/ routerModel type=Our model, Routing=dynamic routing LoRA2026.05 | 2.42 | 2.49 | 15.32 | 9.26 | 7.37 | |
| Qwen3-ASRModel type=Open-source2026.05 | 2.52 | 2.62 | 19.18 | 11.44 | 8.94 | |
| Whisper-L-v3Model type=Open-source2026.05 | 2.85 | 2.97 | 25.68 | 15.65 | 11.79 | |
| Gemini3-FlashModel type=Closed-source2026.05 | 3.1 | 4.27 | 25.99 | 21.86 | 13.81 | |
| Canary-1B-v2Model type=Open-source2026.05 | 3.14 | 3 | 24.88 | 15.56 | 11.65 | |
| Parakeet-v3Model type=Open-source2026.05 | 3.23 | 3.27 | 19.77 | 13.84 | 10.03 | |
| Voxtral-MiniModel type=Open-source2026.05 | 3.5 | 3.51 | 27.54 | 16.45 | 12.75 | |
| Qwen2.5-OmniModel type=Open-source2026.05 | 4.15 | 4.03 | 44.76 | 22.53 | 18.87 | |
| Doubao-LLM ASRModel type=Closed-source2026.05 | 4.86 | 6.99 | 17.23 | 7.85 | 9.23 | |
| GPT-4o-trans.Model type=Closed-source2026.05 | 10.97 | 12.56 | 46.68 | 29.38 | 22.65 |