Meta Muse gets real-time voice, expressive avatars

Meta is turning Muse from a text-based agent into a persistent character, making emotional presence a core product feature rather than a visual gimmick.

• –Shared speech-token streams synchronize voice, lip movement, and gestures in real time.
• –Meta’s system targets 25fps portrait video with roughly 870ms response latency, backed by quantization and optimized inference infrastructure. [Technical overview](https://research.meta.ai/blog/bringing-your-muse-to-life)
• –Hundreds of selectable voices and customizable avatars give Muse stronger identity and retention hooks than conventional voice assistants.
• –The rollout is consumer-focused today, but the underlying realtime avatar stack could eventually power games, virtual characters, education, and customer-service agents.
• –Meta still has to prove that expressive avatars improve task completion rather than simply making AI interactions more addictive.