Haibin Wu

Emotion vectors are not monolithic directions: decoupling the generic shift away from neutrality from the emotion-specific residual more than doubles training-free steering precision in frozen TTS backbones.

Non-verbal vocalizations can be effectively modeled in ASR, improving recognition of rare events without sacrificing lexical accuracy.

LLMs can now evaluate audio as well as humans, without task-specific training, thanks to a new instruction-driven framework.