DyaPlex

Full-duplex speech-motion model for dyadic interaction, enabling a digital agent to speak, listen, perceive partner motion, and generate responsive motion in a streaming loop.

DyaPlex is NVIDIA’s full-duplex speech-motion model for dyadic interaction. It brings streaming speech and physical motion into one interactive system, allowing an agent to listen, speak, perceive a partner’s motion, and generate synchronized body-motion responses in real time.

The project highlights the next step from speech-only full-duplex agents toward multimodal embodied interaction, where conversational behavior depends on both audio and partner motion.

Links: Project page · arXiv