Spoken Instruction Following on Human Recordings (test)
71.2LC Win RateVIRBA
Evaluation Results
| Method | Links | |
|---|---|---|
| VIRBABackbone=Qwen2.5-Omni2025.08 | 71.2 | |
| VIRBABackbone=Qwen2-Audio2025.08 | 68 | |
| Step-Audio-R12025.08 | 65.1 | |
| Qwen2.5-Omni2025.08 | 64.4 | |
| Single-view RL2025.08 | 59 | |
| TTS-SFT2025.08 | 55.4 | |
| Qwen2-Audio base2025.08 | 43.1 |