GNM Voice Lab

All projects

Apps & Tools

A local, browser-based talking avatar powered by Google GNM and WebGPU.

byMajid Manzarpour

PlayableOpen source

GNM Voice Lab media is blocked

Allow external media to connect to the provider and play this content.

Description

An interactive Three.js avatar rig that performs a full voice-to-voice loop locally in the browser. It integrates Google's Generalized Neural Mesh (GNM) for 3D facial articulation with local WebGPU models including Gemma for reasoning, Moonshine for transcription, and Kokoro for speech synthesis.

Descriptions, tags, and model credits may be AI-generated or inferred from public sources and can be incomplete or wrong. Learn more.

Details

Why this is interesting: It demonstrates a remarkably complex real-time interaction loop (listen/think/speak/animate) running entirely locally…
Attributed to: @majidmanzarpour
Built with: GPT-5.6 Sol North Mini Code Grok Voice Think Fast 1.0 GPT Image 2 Codex Gemma 4 E2B/E4B Moonshine Tiny q8 Kokoro Timestamped fp32
Links: avatartalking.netlify.app
View tweet
View GitHub repo
Tags: #avatar #webgpu #threejs #lip-sync

Discovered: X
First posted: X · Jul 18, 2026
Listed: Added 39d ago
Build: Strong
Models: Strong

GNM Voice Lab

Description

Timeline

GNM Voice Lab

Description

Timeline