🎙️ NovaVoice — Model-Agnostic AI Smart Speaker

The open AI speaker that works with ANY model. Switch between OpenAI, Anthropic, Gemini, Llama (local), or any future model with one click.

✨ Features

• 🔌 Model-Agnostic — Works with OpenAI, Anthropic, Groq, Ollama, Hermes Agent, OpenRouter
• 🔒 Privacy-First — Local wake word, physical mic mute, optional cloud
• 🎤 Voice Control — Wake word → Speak → AI responds naturally
• 🏠 Smart Home — MQTT + Home Assistant integration
• 🌐 Web Dashboard — Configure providers, view stats, chat via browser
• 🔧 Open Platform — Open-source firmware, community skills

🚀 Quick Start

Prerequisites

• macOS (tested on Apple Silicon)
• Python 3.9+
• Microphone + Speaker/Headset

Setup

cd ~/novavoice

# Install dependencies
python3 run.py setup

# Or manually:
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Configure

Edit config/default.yaml:

llm:
  default_provider: "openrouter"  # or openai, anthropic, groq, ollama
  
  providers:
    openrouter:
      api_key: "your-key-here"    # Get free key at openrouter.ai
      model: "google/gemma-4-31b-it:free"

Or set environment variables:

export OPENROUTER_API_KEY="your-key"
export OPENAI_API_KEY="your-key"
export GROQ_API_KEY="your-key"

Run

Text Mode (testing without mic)

python3 run.py text

Voice Mode (with mic)

python3 run.py voice

Web Dashboard

cd ~/novavoice
source .venv/bin/activate
uvicorn web.server:app --host 0.0.0.0 --port 8301

Then open http://localhost:8301

🏗️ Architecture

┌──────────────┐     ┌──────────────┐     ┌──────────────┐
│  Microphone  │────▶│  Whisper STT │────▶│  LLM Router  │
│  (sounddevice)│     │ (faster-whisper)│   │  (any API)   │
└──────────────┘     └──────────────┘     └──────┬───────┘
                                                  │
┌──────────────┐     ┌──────────────┐            │
│   Speaker    │◀────│   Edge TTS   │◀───────────┘
│              │     │  (or Piper)  │
└──────────────┘     └──────────────┘

🎯 Voice Commands

Command	Action
"Switch to OpenAI"	Switch LLM provider
"Use Claude"	Switch to Anthropic
"Use local model"	Switch to Ollama
"Clear history"	Reset conversation
"Status"	Show current provider & stats
"List providers"	Show available providers

📁 Project Structure

novavoice/
├── main.py                 # Main application + voice loop
├── run.py                  # Launcher script
├── requirements.txt        # Dependencies
├── config/
│   └── default.yaml        # Configuration
├── audio/
│   ├── capture.py          # Microphone input (sounddevice)
│   └── playback.py         # Speaker output
├── stt/
│   └── whisper_stt.py      # Speech-to-Text (faster-whisper)
├── llm/
│   └── router.py           # Model-agnostic LLM router
├── tts/
│   └── edge_tts_engine.py  # Text-to-Speech (Edge TTS)
├── wake/
│   └── wake_word.py        # Wake word detection
├── web/
│   ├── server.py           # Web dashboard (FastAPI)
│   └── templates/
│       └── dashboard.html  # Dashboard UI
└── PROJECT_PLAN.md         # Full project plan

🔌 Supported Providers

Provider	Models	Free?
OpenAI	GPT-4o, GPT-4o-mini	Paid
Anthropic	Claude Sonnet, Haiku	Paid
Groq	Llama 3.3 70B, Mixtral	Free tier
Ollama	Any local model	Free
Hermes	Hermes Agent	Free
OpenRouter	100+ models	Free tier

🛠️ Hardware (Coming Soon)

See PROJECT_PLAN.md for full hardware design:

• ESP32-S3 based
• 2x MEMS microphone array
• 3W speaker with I2S amp
• WS2812B LED ring
• Target BOM: <$70 at 100 units

📄 License

MIT License — Open source, use however you want.