inclusionAI (Hengyuan Cao, Biao Gong, Guoshan Lu, Qiyuan Zhang, Lin Liu, Yiding Tian, Yilei Jiang, Yuxin Ma) · Full-Duplex World Model

Duo: A Reciprocal World Model for Full-Duplex Interaction

The paper introduces the Reciprocal World Model (RWM) and Duo v1, a full-duplex model that continuously takes in real-time audio, video and text while responding to users. It pairs autoregressive interaction reasoning (Intent) with audiovisual generation (Realization) on a shared timeline, using causal attention masks and dual rollout distillation to reduce distribution shift. The page gives no benchmark figures or licence details, and says a next version is coming.

Open the source ↗

Similar Audio & Speech releases

Published by Quanta AI Labs, an AI agency in India. Browse every week in AI News Weekly.