Automatic Speech Recognition on Noisy LibriSpeech clean (test)
7.43WER (-5dB)Qwen3-Omni-30B-A3B-Instruct
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Qwen3-Omni-30B-A3B-Instruct2026.07 | 7.43 | 2.75 | 1.74 | 1.49 | 1.34 | 1.28 | 1.26 | 2.47 | |
| Voxtral-Small-24B-25072026.07 | 8.39 | 3.68 | 2.2 | 1.87 | 1.61 | 1.58 | 1.59 | 2.99 | |
| Qwen3-Omni-30B-A3B-Thinking2026.07 | 8.63 | 3.66 | 2.29 | 1.95 | 1.72 | 1.69 | 1.64 | 3.08 | |
| Whisper-large-v32026.07 | 9.3 | 4.24 | 2.72 | 2.27 | 2.15 | 1.98 | 1.97 | 3.52 | |
| Audex 30B-A3Bprompts=fixed, sampling=greedy2026.07 | 9.37 | 3.26 | 2.01 | 1.67 | 1.41 | 1.37 | 1.34 | 2.92 | |
| Step-Audio-R1.1 33B2026.07 | 9.74 | 4.05 | 2.74 | 3.55 | 1.72 | 1.69 | 1.61 | 3.59 | |
| Kimi-Audio-Instruct2026.07 | 9.81 | 3.86 | 2.06 | 1.6 | 1.47 | 1.35 | 1.33 | 3.07 | |
| Audex 2Bprompts=fixed, sampling=greedy2026.07 | 10.6 | 4.02 | 2.41 | 1.97 | 1.66 | 1.6 | 1.63 | 3.41 | |
| Canary Qwen 2.5B2026.07 | 11.06 | 4.73 | 2.5 | 2.91 | 1.64 | 1.61 | 1.61 | 3.58 | |
| Parakeet-TDT-0.6B-v32026.07 | 12.21 | 4.82 | 2.62 | 2.15 | 1.96 | 1.92 | 1.92 | 3.94 | |
| Canary-1B-v22026.07 | 19.38 | 5.08 | 2.8 | 2.29 | 2.01 | 2.16 | 2.18 | 5.13 |