This project provides a fully local, GPT-like voice assistant designed for Raspberry Pi, enabling private, offline voice command interaction. It targets hobbyists and users seeking a self-contained AI assistant without cloud dependencies.

How It Works

The assistant records audio via a pushbutton and USB microphone, transcribes it using Whisper, and prompts a local TinyLlama-1.1B LLM packaged as a llamafile. The LLM's text response is then synthesized into speech using eSpeak and played through a speaker. This approach ensures complete data privacy and offline operation.

Quick Start & Requirements

• Install: Ubuntu Server 22.04.3 LTS (64-bit) on SD card, then run sudo apt update, sudo apt install ffmpeg espeak python3-pip python3-pyaudio, pip3 install openai openai-whisper RPi.GPIO pyaudio, clone repo, download and chmod the llamafile.
• Prerequisites: Raspberry Pi 4 (or Pi 5), USB microphone, speaker, pushbutton, SD card.
• Setup: Requires flashing OS, installing packages, downloading LLM model.
• Docs: Demo Video

Highlighted Details

• Runs 100% locally on a Raspberry Pi 4.
• Utilizes Whisper for speech-to-text and TinyLlama-1.1B via llamafile.
• Requires Ubuntu Server 22.04.3 LTS; Raspberry Pi OS is incompatible without custom kernel workarounds.

Maintenance & Community

• Project by Nick A. Bild. No community links or roadmap provided.

Licensing & Compatibility

• No explicit license mentioned in the README.
• Compatibility is limited to Ubuntu Server on Raspberry Pi due to llamafile requirements.

Limitations & Caveats

Processing can be slow, with brief requests taking up to 15 seconds on a Pi 4. The project is presented as "World's Easiest," but setup involves OS flashing and command-line operations.