HeyGen Open-Sources MIT Demo: GPT-Live Voice Model Driving a Real-Time Teaching Avatar
HeyGen · x · 2026-09-11
HeyGen open-sourced (MIT) liveavatar-gpt-live-demos, a minimal reference integration of OpenAI's GPT-Live full-duplex speech-to-speech model driving a HeyGen LiveAvatar realtime voice agent, with tool calls rendered as on-screen animated overlays. The out-of-box demo is a Japanese tutor that teaches words aloud with on-screen term cards and periodic recaps rendered from the server's session record rather than model memory. Personas are two markdown files; the repo is deliberately barebones so the wiring itself is the thing you read.
More from Apps
- Gather launches research API for AI agents, already used by 60+ enterprises — ycombinator · 2026-10-27
- Intent mobile v1.1.7 adds file attachments, workspace browsing and slash skill suggestions — LukeW · 2026-10-27
- AI Overviews on 80% of queries will make Google Search Console data a mess — lilyraynyc · 2026-10-27
- Sunlight: free open-source infinite canvas window manager for Mac — msg · 2026-10-27
- Law professor uses Opus 5.5 to produce a Like Company v Google copyright explainer video — technollama · 2026-10-27
- Open-source Clairic Studio turns pasted text into typed data and working code in 5 languages — dfinke · 2026-10-27