local-talking-llm

Why we included this project

The fastest way to tell whether a local model can actually hold a conversation is to talk to it, and this project turns that test into a complete two-way setup. It chains Whisper for speech recognition, Ollama as the default LLM backend, and Chatterbox for text-to-speech, so a spoken question comes back as a spoken answer without any cloud calls. The TTS side clones a voice from a short audio sample and adds emotion control plus watermarked output for anyone who does not want a flat robotic tone. If you are building a Jarvis-style assistant or want to bolt speech onto a model you already run locally, the Python code is easy to read and swap over to another provider like MiniMax.