Automatic Speech Translation on Fleurs (Multi-directional, BLEU & COMET)
43.9BLEU Score (DE)Voxtral-Small-24B-2507
Evaluation results
| Method | Links | ||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Voxtral-Small-24B-2507Model size=24B2026.07 | 43.9 | 89 | 34.8 | 87.5 | 42.6 | 88.6 | 35.2 | 88 | 48.7 | 89.2 | 37.2 | 86.6 | 26.6 | 85.7 | 41 | 88.5 | |
| Qwen3-Omni-30B-A3B-InstructModel size=30B-A3B, Variant=Instruct2026.07 | 41.5 | 88.5 | 28.6 | 86.7 | 38.1 | 87.9 | 28.6 | 87 | 43.3 | 88.3 | 32.4 | 86 | 28.1 | 87.6 | 34.4 | 87.4 | |
| Qwen3-Omni-30B-A3B-ThinkingModel size=30B-A3B, Variant=Thinking2026.07 | 40.7 | 88.7 | 28 | 86.6 | 37.6 | 88.3 | 28.2 | 87 | 42.7 | 88.5 | 32.3 | 86.1 | 27.6 | 87.9 | 33.9 | 87.6 | |
| Audex 30B-A3BModel size=30B-A3B2026.07 | 40 | 88 | 28.5 | 86.1 | 38 | 87.1 | 29 | 86.8 | 43.4 | 88 | 32.8 | 85.3 | 26.3 | 86.9 | 34 | 86.9 | |
| Step-Audio-R1.1 33BModel size=33B2026.07 | 38 | 88.2 | 26.8 | 86.4 | 36.2 | 87.3 | 26.7 | 86.7 | 41.7 | 87.7 | 30.2 | 85.1 | 26.4 | 86.8 | 32.3 | 86.9 | |
| Audex 2BModel size=2B2026.07 | 37.3 | 86.8 | 35.2 | 85.8 | 26 | 85.6 | 26.5 | 85.3 | 41.3 | 87 | 29.4 | 83.5 | 23.4 | 85.5 | 31.2 | 85.7 | |
| Seamless-m4t-v2-large2026.07 | 36.8 | 86 | 25.7 | 84.2 | 33.9 | 85.2 | 26.8 | 85 | 37.4 | 85 | 30.2 | 83.9 | 22.8 | 83.5 | 30.5 | 84.7 | |
| Canary-1B-v2Model size=1B2026.07 | 35.2 | 85.4 | 25.2 | 83.3 | 33.7 | 85.2 | 25.1 | 84.2 | 38.6 | 85.4 | 27 | 81.6 | — | — | — | — | |
| Whisper-large-v32026.07 | 33.6 | 84.8 | 22.5 | 83.5 | 31.1 | 84.4 | 23 | 84 | 36.8 | 85.8 | 26.5 | 83.3 | 19.2 | 81.7 | 27.5 | 83.9 |