frank/weebo
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
A real-time speech-to-speech chatbot powered by Whisper Small, Llama 3.2, and Kokoro-82M.
2 commits 1 branch 0 tags 9.2 MiB
  • Python 100%
Find a file
Exact
Amanvir Parhar b0acd79ed9
 Create LICENSE
2025-01-15 17:49:53 -08:00
.gitignore initial commit 2025-01-15 17:11:46 -08:00
LICENSE Create LICENSE 2025-01-15 17:49:53 -08:00
main.py initial commit 2025-01-15 17:11:46 -08:00
README.md initial commit 2025-01-15 17:11:46 -08:00
requirements.txt initial commit 2025-01-15 17:11:46 -08:00
voices.json initial commit 2025-01-15 17:11:46 -08:00

README.md

Weebo

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

Works on Apple Silicon.

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)
  • Pull the llama3.2 model using Ollama

Usage

Run the chatbot:

python weebo.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.