Speech-to-text translation on CoVoST
38.2Average BLEUGemma 4 E4B
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Gemma 4 E4BParams=305M, Size=87 MB, S2TT prompt=transcribe then translate2026.07 | 38.2 | 25.5 | 42 | 41 | 44.8 | 43 | 49.4 | 21.9 | — | |
| VIRBABackbone=Qwen2.5-Omni2025.08 | 36.1 | — | — | — | — | — | — | — | — | |
| Kimi-Audio2025.08 | 35.4 | — | — | — | — | — | — | — | — | |
| Gemma 4 E2BParams=305M, Size=87 MB, S2TT prompt=transcribe then translate2026.07 | 35.4 | 21.4 | 39.2 | 39.2 | 43.2 | 40.8 | 46.4 | 17.9 | — | |
| Qwen2.5-Omni2025.08 | 35 | — | — | — | — | — | — | — | — | |
| Gemma 3n E4BParams=680M, Size=390 MB, S2TT prompt=transcribe then translate2026.07 | 34.7 | 22.3 | 39.1 | 38.4 | 41.8 | 40.4 | 43.7 | 17.4 | — | |
| Step-Audio-R12025.08 | 34.2 | — | — | — | — | — | — | — | — | |
| AF32025.08 | 33.9 | — | — | — | — | — | — | — | — | |
| VIRBABackbone=Qwen2-Audio2025.08 | 33.7 | — | — | — | — | — | — | — | — | |
| Group-DPO2025.08 | 33 | — | — | — | — | — | — | — | — | |
| SeamlessM4T-L V22025.08 | 32.8 | — | — | — | — | — | — | — | — | |
| Gemma 3n E2BParams=680M, Size=390 MB, S2TT prompt=transcribe then translate2026.07 | 31.6 | 17.7 | 36.5 | 35.7 | 39.9 | 38.5 | 39.2 | 13.9 | — | |
| Qwen2-Audiovariant=base2025.08 | 30.3 | — | — | — | — | — | — | — | — | |
| Whisper-large V22025.08 | 28 | — | — | — | — | — | — | — | — |