A fully local, offline-capable voice assistant that lets you speak a question and get a spoken response — powered by Whisper, LLaMA2 via Ollama, and pyttsx3. Built with a modern and professional Gradio interface.

• 🎤 Voice Input (Record your question via mic)
• 🧠 Whisper (STT) + LLaMA2 (via Ollama) for question answering
• 🔊 Text-to-Speech using pyttsx3
• 💬 Beautiful chat-style UI with transcript and assistant response
• 🖥️ Runs 100% locally – No API keys needed

• Python 3.8+
• Windows/macOS/Linux
• 8GB+ RAM (recommended)
• For GPU support (optional): NVIDIA GPU with CUDA

git clone https://github.com/your-username/ai-voice-assistant.git
cd ai-voice-assistant

python -m venv venv
source venv/bin/activate      # On Windows: venv\Scripts\activate

pip install -r requirements.txt

💡 If you get CUDA or cuDNN errors, change Whisper’s compute_type="int8_float32" and device="cpu" in whisper_module.py.

1. Download and install Ollama: 👉 https://ollama.com/download

2. Pull a local LLaMA2 model:

   ollama pull llama2

3. Keep Ollama running in the background:

   ollama run llama2

python app.py

• Click "🎧 Start Talking"
• Speak your question (e.g., "What is the capital of Japan?")
• Watch the transcript + assistant response appear
• Listen to the assistant speak back

Feel free to add any issues you find.

MIT License. Use freely for personal, academic, or demo purposes.

Deshan Senanayake