Emotion-Aware Personalized Co-Speech Motion Generation

Co-speech motion generation is a challenging task requiring alignment between motion, speech rhythm, and semantics. To improve adaptability, style control mechanisms, including emotion perception and personalized motion generation, are needed. We propose a hybrid emotion-aware personalized co-speech motion generation system, consisting of an emotion-aware module based on LLM and a rhythm motion generation module based on diffusion, ensuring consistency between motion and speech emotion. Additionally, we introduce a style transfer module to adapt the generated motions to the speaker's style. Experimental results show the system generates stylized and emotionally consistent motions, enhancing the realism and stylization of virtual humans.