Speech To Speech: Build voice agents with open-source models
A low-latency, fully modular voice-agent pipeline:
VAD -> STT -> LLM -> TTS
, exposed through the
core OpenAI Realtime GA event set over WebSocket and WebRTC
. Every component is swappable. The LLM slot speaks OpenAI-compatible protocols, so you can point it at a hosted provider, at
HF Inference Providers
, or at a vLLM or llama.cpp server on your own hardware for a fully local, fully open stack.
This pipeline runs in production as the conversation backend for thousands of
Reachy Mini
robots.
Quickstart
Choose where the language model should run. All three configurations use local Parakeet TDT speech recognition and Qwen3-TTS speech output
by default
. You can change the STT, LLM, and TTS models and backends; see
Supported compon (EN)

---
**📖 中文解读**
以上内容由AI翻译自英文原文，可能存在不准确之处。建议阅读[原文](https://github.com/huggingface/speech-to-speech)获取最准确的信息。

---
🔗 **原文链接**: [huggingface/speech-to-speech (⭐ 26 stars today)](https://github.com/huggingface/speech-to-speech)
🏷️ **转载来源**: GitHub Trending
> 本文由小九AI技术站翻译整理，内容版权归原作者所有。

---
🐾 **小九锐评**

多模态正在逼近实用门槛。如果你想做产品级落地，这篇文章值得读。

你对这个话题有什么看法？欢迎在评论区讨论 💬

> _转载自 GitHub Trending，内容版权归原作者所有_

---
⏱️ 2026-09-10 22:01