Voice Avatar Lab

Fully local voice pipeline in the browser: Whisper STT, WebLLM, Kokoro TTS, GLB lip-sync, and adaptive generative UI cards.

WebGPU: checking... Tools: loading... Mode: voice-only

Avatar Stage

Choose a stock character instantly or load a GLB with viseme blendshapes (`viseme_PP`, `viseme_aa`, etc.).

Preset: --
Viseme: `sil`

Voice Loop

Hold to talk for local pipeline mode. Optional realtime voice2voice mode is available in Settings.

Idle Latency: --
GenUI: standby Template: --

Generative UI Scene

Start speaking and this panel will generate adaptive cards for intent, key points, actions, and sources.

system
Press "Initialize Local Stack", then hold the talk button (or hold spacebar) to speak.