VoxCore

Local voice assistant pipeline — STT → LLM → TTS, 100% offline.

Stack : faster-whisper + Qwen3.6 + Voxtral (ou Supertonic/Piper)

Installation :

pip install voxcore
voxcore --config config.yaml

Usage :

# Microphone → réponse vocale
voxcore live

# Audio file → texte
voxcore stt input.wav

# Texte → audio
voxcore tts "Hello world" -o output.mp3

Config : config.yaml dans le dossier courant ou ~/.config/voxcore/config.yaml

Architecture

• stt/ — Speech-to-Text (faster-whisper)
• llm/ — Local LLM (llama.cpp, vLLM, LM Studio)
• tts/ — Text-to-Speech (Voxtral, Piper, Supertonic)
• pipeline/ — Orchestration complète

Requirements

• Python 3.10+
• GPU NVIDIA recommandé (CUDA 12+)
• RAM : 16GB minimum

License

MIT