Hugging Face's open-source voice-agent pipeline runs in production on thousands of robots
adnan_hashmi · x · 2026-09-16
Hugging Face's speech-to-speech repo is a fully modular, low-latency voice-agent pipeline: VAD → STT → LLM → TTS, exposed via the OpenAI Realtime GA event set over WebSocket and WebRTC.
Key points:
- Every component is swappable: the LLM slot speaks OpenAI-compatible protocols, so it can target hosted providers, HF Inference Providers, or a local vLLM/llama.cpp server for a fully local, fully open stack
- Production-proven: it runs as the conversation backend for thousands of Reachy Mini robots
- 13.2k stars and 1.7k forks on GitHub
- Ships with Docker support, an ARM64 Dockerfile, and a quickstart guide
More from coding & agent
- When an LLM plans an executable agent DAG, where do you draw the trust boundary? — Repulsive_Sugar_5252 · 2026-09-16
- "Never again": user rants Nemotron 3 Ultra can't find .github and stalls for 90 minutes — dwerps · 2026-09-16
- Personal AI's real test: does your judgment survive after you close the window? — sujingshen · 2026-09-16
- Coding agent so fears AWS costs it wrote a test banning NAT gateways — mimi10v3 · 2026-09-16
- Muse Code raises contributor tier limits and resets usage for all subscribers — giffmana · 2026-09-16
- OpenAI reportedly orchestrates 10k internal agents without quality loss, while users' sub-agents 'just burn tokens' — RexDouglass · 2026-09-16