List your company

Spoken Instruction Following on TTS Unseen (test)

77LC Win RateVIRBA
47.67255.28662.970.514Aug 25, 2025

Evaluation results

MethodLinks
VIRBABackbone=Qwen2.5-Omni2025.0877
VIRBABackbone=Qwen2-Audio2025.0873.8
Step-Audio-R12025.0871.6
Qwen2.5-Omni2025.0870.5
Single-view RL2025.0866.2
TTS-SFT2025.0861
Qwen2-Audio base2025.0848.8