Audio-driven lipsync
Real-time test of the three.ws viseme analyser
(src/lip-sync-analyser.js). Hit Start,
allow mic access, and speak — the analyser hands per-viseme weights
to the avatar's morph targets every frame. This is the exact pipeline
that drives ElevenLabs-streamed TTS in the live chat, just sourced
from your mic so you can verify the mouth shapes against your speech.
The avatar's
viseme_aa / E / O / SS / PP … morph
targets are driven by frequency-band energy from the mic. Your
audio never leaves the browser — analysis is all
AnalyserNode + requestAnimationFrame.