OllamaGTTS - Voice Assistant for Ollama

A lightweight voice assistant that uses Ollama for AI responses and Google Text-to-Speech (gTTS) for voice output, featuring real-time voice interaction, conversation memory, and audio processing optimizations.

Key Features

• Voice Interaction

  • Real-time speech detection using Silero VAD
  • Whisper-based transcription (faster-whisper)
  • Interruptible speech playback
  • Background audio processing

• Enhanced Audio

  • Google TTS with natural chunking
  • FFmpeg-accelerated playback (1.15x speed - Optional)
  • Audio queue prioritization system
  • Automatic temp file cleanup

• Conversation Management

  • Persistent conversation history (JSON)
  • Context-aware prompting
  • Model-specific system prompts
  • Configurable history length

• Technical Features

  • GPU acceleration support (CUDA)
  • Multi-threaded audio processing
  • Cross-platform compatibility
  • Model selection interface

Requirements

• Python 3.7+
• Ollama installed and running locally
• Internet connection (for Google TTS service)
• FFmpeg (optional - for audio speed adjustment)

Installation

1. Clone the repository

git clone https://github.com/ExoFi-Labs/OllamaGTTS.git
cd OllamaGTTS

2. Create a virtual environment (optional but recommended)

python -m venv venv

Activate the virtual environment:

• On Windows:

  venv\Scripts\activate

• On macOS/Linux:

  source venv/bin/activate

3. Install required Python packages

pip install -r requirements.txt

[!NOTE] To install pyaudio in macOS environment, please install portaudio first.

xcode-select --install
brew install portaudio

4. Install Ollama

If you haven't already installed Ollama, follow the instructions at Ollama's official website.

Make sure you have at least one model downloaded using:

ollama pull llama3.3

or any other model of your choice.

5. (Optional) Install FFmpeg for audio speed adjustment

FFmpeg is used to adjust the speed of audio playback. The application works without FFmpeg, but audio will play at normal speed.

• Windows: Download from FFmpeg's official website and add to your PATH
• macOS: Install using Homebrew: brew install ffmpeg
• Ubuntu/Debian: Install using apt: sudo apt install ffmpeg
• Other Linux: Use your distribution's package manager
• Untested on Mac/Ubuntu

Usage

1. Run the application:

python ollama_gttsg.py

3. Select a model from the list of available models
4. Enter a system message or press Enter to use the model's default
5. Start your conversation with the assistant

Commands

• Type your message and press Enter to send
• Type exit or quit to end the conversation

How It Works

1. The application connects to your local Ollama instance and lists available models
2. When you send a message, it gets streamed to the selected Ollama model
3. As responses are received, the text is chunked at natural pause points
4. Each chunk is converted to speech using Google's TTS service
5. Audio is played back in the correct order with speed adjustment (if FFmpeg is available)
6. Conversation history is stored for future context

Customization

Changing the Voice

To change the TTS voice language, modify the lang parameter in the create_and_queue_audio function. Default is English ('en').

Adjusting Speech Speed

If you have FFmpeg installed, you can change the speech speed by modifying the speed_factor value in the create_and_queue_audio function. The default is 1.15 (15% faster than normal).

Audio settings (Modify these values in ollama_gttsg.py:)

vad_threshold = 0.5 # Speech detection sensitivity (0.3-0.7) silence_duration = 1.0 # Seconds of silence to end speech speed_factor = 1.15 # Playback speed multiplier

Conversation settings

max_history = 10 # Number of exchanges to remember

Troubleshooting

No audio output

• Make sure your system's audio is not muted
• Check that pygame is properly installed
• Try restarting the application

Models not showing up

• Make sure you've downloaded at least one model using ollama pull

FFmpeg not found

• If you want audio speed adjustment, make sure FFmpeg is installed and available in your system PATH
• Without FFmpeg, the application will still work but will use normal speed audio

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

This project is licensed under the MIT License - see the LICENSE file for details.