Brief

Researchers have developed a new method called "semantic motion anchors" to improve the understanding and generation of co-speech gestures. This approach bridges the gap between spoken language and physical motion by creating natural-language abstractions of gestures that capture both their form and communicative intent. By discretizing gesture movements and verbalizing them, the system provides auxiliary supervision that enhances retrieval accuracy and leads to more semantically meaningful gesture generation. AI

Researchers have developed PersonaGesture, a new diffusion-based system designed to personalize co-speech gestures for unseen speakers. The system takes target speech and a single motion clip from a new individual to generate gestures that match the utterance while preserving the speaker's unique style. PersonaGesture utilizes Adaptive Style Infusion and Implicit Distribution Rectification to effectively separate speaker identity from utterance-specific motion, improving personalization compared to existing methods. AI