A model renders a listening human face in real time

Claim

Tavus launched Phoenix-4, the first real-time human rendering model unifying emotional expression, active listening and facial motion, while ElevenLabs' Scribe v2 reached a 2.3% speech-to-text error rate and Google's Lyria 3 generated music from images.

Reported metric: 2.3% error

Reporting date: 2026-02-19 (not necessarily the event date)

Sources

Attribution and reporting linked by this record; links do not imply independent verification.

• Welcome to February 19, 2026 (theinnermostloop.substack.com)