Full-duplex is a mode of communication in which both parties can transmit and receive at the same time. A telephone call, where two people can talk and listen simultaneously, is full-duplex, in contrast to half-duplex, where only one side can send at a time. The term originates in telecommunications and networking.
In AI, the term describes the architecture of spoken dialogue models. Unlike turn-based systems that wait for the user to finish before responding, a full-duplex speech model generates its own speech while continuously listening to the other party, which makes interruptions, backchannel responses and natural conversational timing possible. It is mainly adopted in end-to-end speech models that process audio directly rather than through text.