Meta Gives AI Assistant Muse Real-Time Voice and Video Capabilities
Meta CEO Mark Zuckerberg announced that the Muse AI assistant now supports real-time voice conversations and expressive avatars. Initially rolling out to users in the United States, the feature is available across mobile, web, and Mac applications. This update transforms Muse into a more natural and synchronous conversational agent, enabling low-latency face-to-face interactions. It represents a significant step forward in making AI assistants feel more embodied and responsive during complex tasks. Powered by Muse Realtime Voice and Muse Realtime Avatar models, the system streams vertical video at 448x768 resolution and 25 frames per second. It achieves an end-to-end response latency of approximately 870 milliseconds from the moment a user finishes speaking.
## BACKGROUND
Conversational AI assistants traditionally relied on sequential text-to-speech pipelines, which resulted in unnatural delays and robotic interactions. Recent breakthroughs in streaming multi-modal models allow AI systems to process audio and visual tokens simultaneously, creating fluid real-time communication experiences.