Speech-to-Text Translation on CoVoST2 en-zh
55.22BLEUAudio-Interaction
Evaluation Results
| Method | Links | |
|---|---|---|
| Audio-InteractionSize=3B2026.06 | 55.22 | |
| MiniCPM-o 4.5Size=9B2026.04 | 49.9 | |
| Qwen3-OmniSize=30B-A3B2026.04 | 46.6 | |
| Phi-4-multimodalSize=5.6B2026.06 | 46.3 | |
| Qwen2-AudioSize=7B2026.06 | 45.2 | |
| Qwen-Audiozero-shot=true2023.11 | 41.5 | |
| Qwen2.5-OmniTrain audio duration(h)=>1000k2026.04 | 41.4 | |
| Qwen2.5-OmniSize=7B2026.06 | 41.4 | |
| Qwen2.5-OmniSize=3B2026.06 | 39.5 | |
| SLAMTrain audio duration(h)=1.8k2026.04 | 37.34 | |
| Kimi-AudioSize=9B2026.04 | 36.6 | |
| TASUTrain audio duration(h)=02026.04 | 33.35 | |
| TASU2Train audio duration(h)=02026.04 | 33.08 |