Hugging Face's speech-to-speech repo gains 1,342 stars in a day

Hugging Face's speech-to-speech repository picked up 1,342 stars in a single day, per its GitHub page. That's about twice the next-fastest Python project's daily gain, and it pushed the total past 9,740 stars.

The repo already runs in production on Reachy Mini robots. That's real infrastructure, not a demo racking up stars for a slick video.

The system chains four stages: voice activity detection, speech-to-text, an LLM, and text-to-speech, wrapped behind a WebSocket API compatible with OpenAI's Realtime API.

Defaults are Silero VAD v5 for detection, Parakeet TDT for transcription, and Qwen3-TTS for speech output. Each stage swaps out, though: Whisper or Paraformer can replace Parakeet, and Kokoro, Pocket TTS, ChatTTS, or MMS TTS can replace Qwen3-TTS. The LLM connects through an OpenAI-compatible endpoint, plain Transformers, or MLX-LM for Apple Silicon.

None of this is new, either. The repo dates to August 2024, and per GitHub's trending page, the July 31 spike reads as a resurgence, not a launch.