Speech robustness on Speech-view Consistency
96.9ConsistencyVIRBA
Evaluation results
| Method | Links | |
|---|---|---|
| VIRBABackbone=Qwen2.5-Omni2025.08 | 96.9 | |
| VIRBABackbone=Qwen2-Audio2025.08 | 96.2 | |
| Step-Audio-R12025.08 | 94.6 | |
| Group-DPO2025.08 | 94.5 | |
| Qwen2.5-Omni2025.08 | 94.1 | |
| Kimi-Audio2025.08 | 93.8 | |
| AF3Full Name=Audio Flamingo 32025.08 | 93.1 | |
| Single-view RL2025.08 | 91.7 | |
| GLM-4-Voice2025.08 | 91.2 | |
| TTS-SFT2025.08 | 89.3 | |
| Qwen2-Audio base2025.08 | 82.6 |