Meta introduced Muse Realtime Voice and Muse Realtime Avatar on November 5, enabling live voice conversations and responsive avatars.
On November 5, Meta began rolling out Muse Realtime Voice and Muse Realtime Avatar in the U.S. Muse app for mobile, web and Mac, letting users speak with the AI agent and interrupt its replies. People can choose voice-only, an avatar beside the chat or a full-screen view; Meta says hundreds of voice styles are available. The avatar reacts while listening and speaking, with expressions and gestures matched to the exchange. Meta’s research describes a streaming system: Muse Realtime Voice produces speech tokens used by both an audio decoder and the avatar model. The avatar generates 448-by-768 portrait video at 25 frames per second, with about 870 milliseconds from the end of a user’s turn to the first synchronized response. On a GB200, Meta reports 12 concurrent video sessions, an eightfold serving-capacity gain over its two-step BF16 baseline. The launch gives Meta a visual, interruptible voice experience to compete with conversational AI assistants, while the initial U.S.-only rollout limits access.