Weebo

A real-time speech-to-speech chatbot powered by Whisper Small, Llama 3.2, and Kokoro-82M.

Works on Apple Silicon.

Learn more here.

Features

• Continuous speech recognition using Whisper MLX
• Natural language responses via Llama
• Real-time text-to-speech synthesis with Kokoro-82M
• Support for different voices
• Streaming response generation

Setup

Download required models:

• kokoro-v0_19.onnx (TTS model): wget https://github.com/thewh1teagle/kokoro-onnx/releases/download/model-files/kokoro-v0_19.onnx
• Pull the llama3.2 model using Ollama: ollama pull llama3.2
• for Mac: brew install espeak-ng
• for Mac: export ESPEAK_DATA_PATH=/opt/homebrew/share/espeak-ng-data

Usage

Run the chatbot:

uv run --python 3.12 --with-requirements requirements.txt main.py

The program will start listening for voice input. Speak naturally and wait for a brief pause - the bot will respond with synthesized speech. Press Ctrl+C to stop.

Alternatively, create an environment and install the requirements:

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python main.py