Skip to content

README.md

Latest commit

 

History

History
97 lines (71 loc) · 2.27 KB

README.md

File metadata and controls

97 lines (71 loc) · 2.27 KB
Raw

Voice Assistant with LLaMA.cpp + Gemma2

Original implementation using LLaMA.cpp and Google's Gemma2 models, optimized for the Jetson Orin Nano platform.

Overview

This directory contains the original voice assistant implementation along with additional demonstration scripts showcasing different AI capabilities.

Demos Included

1. Voice Assistant (assistant.py)

  • Main voice assistant with RAG (Retrieval-Augmented Generation)
  • Uses Whisper for speech-to-text
  • FAISS for document search
  • Gemma2 for response generation
  • Piper TTS for text-to-speech

2. Multi-Agent NPC System (npcservers.py)

  • Simulates a conversation between two AI agents
  • Each agent has a different personality
  • Demonstrates multi-agent communication
  • Uses two instances of Gemma2 running simultaneously

3. Translation Assistant (translate.py)

  • English to Japanese voice translation
  • Real-time speech translation
  • Uses Whisper for transcription
  • Gemma2 for translation
  • Coqui TTS for Japanese speech synthesis

Quick Start

Prerequisites

  • LLaMA.cpp server running with Gemma2 model
  • Piper TTS installed (for voice assistant)
  • Coqui TTS installed (for translation demo)

Installation

pip install -r requirements.txt

Running the Demos

Voice Assistant:

python assistant.py

Multi-Agent NPC System:

python npcservers.py

Translation Assistant:

python translate.py

Configuration

LLaMA.cpp Server

Make sure your LLaMA.cpp server is running on port 8080 with a Gemma2 model loaded.

Audio Devices

Update the find_device() function in each script to match your audio input device.

TTS Paths

Update the Piper TTS path in the scripts to match your installation:

piper_path = "/home/asier/piper/build/piper"

Requirements

  • Python 3.8+
  • LLaMA.cpp server
  • Gemma2 model files
  • Piper TTS (optional)
  • Coqui TTS (for translation demo)

Platform Support

Tested and optimized for:

  • Jetson Orin Nano (ARM64, Linux)

Differences from Gemma3

This implementation uses LLaMA.cpp instead of Ollama, providing:

  • Direct model control
  • Custom server configuration
  • Legacy compatibility
  • Research and development capabilities

License

This project is open source and available under the MIT License.