Automatic Speech Recognition on NOIZEUS
19.8Error Rate (0dB SNR)Mega-ASR
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Mega-ASRModel type=Our model2026.05 | 19.8 | 6.61 | 2.79 | 0.88 | 7.52 | |
| Mega-ASR w/ routerModel type=Our model, Routing=dynamic routing LoRA2026.05 | 19.8 | 6.97 | 3.05 | 1.76 | 7.9 | |
| Qwen3-ASRModel type=Open-source2026.05 | 23.97 | 8.47 | 3.41 | 1.96 | 9.45 | |
| Doubao-LLM ASRModel type=Closed-source2026.05 | 25.78 | 9.51 | 4.96 | 2.87 | 10.78 | |
| Step-Audio-2-miniModel type=Open-source2026.05 | 32.02 | 8.94 | 3.72 | 2.27 | 11.74 | |
| Whisper-L-v3Model type=Open-source2026.05 | 34.71 | 12.55 | 3.93 | 2.17 | 13.34 | |
| Kimi-AudioModel type=Open-source2026.05 | 38.33 | 11.36 | 4.34 | 2.27 | 14.08 | |
| Canary-1B-v2Model type=Open-source2026.05 | 38.53 | 12.76 | 6.56 | 3.77 | 15.41 | |
| Parakeet-v3Model type=Open-source2026.05 | 38.95 | 14.67 | 5.99 | 3.15 | 15.69 | |
| Voxtral-MiniModel type=Open-source2026.05 | 41.06 | 15.8 | 4.85 | 2.94 | 16.16 | |
| Qwen2.5-OmniModel type=Open-source2026.05 | 54.91 | 17.72 | 3.2 | 0.88 | 19.18 | |
| Gemini3-FlashModel type=Closed-source2026.05 | 55.78 | 24.48 | 18.49 | 8.52 | 26.82 | |
| GPT-4o-trans.Model type=Closed-source2026.05 | 62.4 | 20.56 | 6.15 | 2.64 | 22.94 |