Spoken Instruction Following on Code-switch (test)
60.3LC Win RateVIRBA
Evaluation Results
| Method | Links | |
|---|---|---|
| VIRBABackbone=Qwen2.5-Omni2025.08 | 60.3 | |
| VIRBABackbone=Qwen2-Audio2025.08 | 56.8 | |
| Step-Audio-R12025.08 | 53.4 | |
| Qwen2.5-Omni2025.08 | 52.1 | |
| Single-view RL2025.08 | 47.6 | |
| TTS-SFT2025.08 | 44.3 | |
| Qwen2-Audio base2025.08 | 35 |