Nova Voice Assistant

A smart, local-first voice assistant that listens for the wake word "Hey Nova," transcribes voice input using OpenAI Whisper, responds with dynamic memory-driven conversations via GPT, and stores facts in Supabase and local storage.

---

Features

• Wake-word detection with Porcupine
• Long-term memory saved in Supabase and user_memory.json
• Responses powered by OpenAI GPT
• Voice transcription using Whisper
• Text-to-speech response via pyttsx3
• Offline fallback with JSON-based memory store

---

Requirements

• Python 3.10 or newer
• OpenAI API key
• Supabase project + API credentials
• Porcupine .ppn wake-word model (included)

---

Setup Instructions

1. Clone this repo

git clone https://github.com/yourusername/nova-voice-assistant.git
cd nova-voice-assistant

2. Create and activate a virtual env

python -m venv agentEnv
.\agentEnv\Scripts\activate

3. Install dependencies

pip install -r requirements.txt

Make sure FFmpeg is installed and added to System PATH Add to system environment variables

Download from: https://www.gyan.dev/ffmpeg/builds/
Add the /bin folder path (e.g., C:\ffmpeg\bin) to your system environment variables.

4. Add environment variables

Create a .env file and add

OPENAI_API_KEY=your-openai-key
SUPABASE_URL=https://your-project.supabase.co
SUPABASE_KEY=your-anon-or-service-role-key

5. Start the assistant

python assistant.py

If successful, you should see:

🔊 Porcupine wake word engine listening for 'Hey Nova'...
🔁 Press [Enter] to ask something or type 'exit' to quit:

---

A Testing Workflow Should Be

1. Say "Hey Nova"
2. Press [ENTER]
3. Speak to Nova
4. Nova will transcribe, respond using GPT, and speak back the reply.