Gemini Live × LAM A2E (real-time test)

Gemini Live (24 kHz PCM) → resampled 16 kHz → LAM streaming A2E (CUDA L4) → blendshape frames over WebSocket → 3DGS head.

Checking…

Audio: 0.00sFrames: 0offline

Time to first lip-sync—

waiting for first turn…

A2E per 200 ms chunk

—

Mouth lag behind audio

0 ms

Server log will appear here…

How to run the bridge

# In a separate terminal:
bash scripts/gemini-live-test.sh
# (uses GEMINI_API_KEY from .env / .env.local — same key the rest of the app uses)
# Server listens on ws://localhost:7777