GPT Phone

GPT Phone is a Python-based project that simulates a phone call with various characters using OpenAI's GPT-3.5 and ElevenLabs for text-to-speech. The project is designed to run on a Raspberry Pi with a physical phone interface. It also interfaces via GPIO pins to a Bell 304 series telephone's hardware including the rotary dial switch, hook switch, handset speaker, and handset microphone.

Installation

1. Create a virtual environment:

   python3 -m venv gpt-phone
   source gpt-phone/bin/activate

2. Install system dependencies:

   sudo apt install portaudio19-dev pipewire-audio-client-libraries

3. Install Python dependencies:

   pip3 install -r requirements.txt

Hardware

1. Connect the hook switch to GPIO14 (or change the pin definition in gpt-phone.py) and ground
2. Connect the rotary dial switch to GPIO15 (or change the pin definition in gpt-phone.py) and ground
3. Connect your audio device's microphone to the phone's microphone circuit
4. Connect your audio device's speaker to the phone's speaker circuit

Configuration

• Personality: Personalize the system roles, greetings, etc in roles.py file.
• Voices: Set the voice IDs for ElevenLabs in roles.py array.
• Dialtone Files: Optionally, change the dialtone files in roles.py array.
• **

Usage

1. Set the environment variables OPENAI_API_KEY and ELEVENLABS_KEY:

   export OPENAI_API_KEY="your_openai_api_key"
   export ELEVENLABS_KEY="your_elevenlabs_api_key"
   export TEST_DIGIT=1 # Optional. Forces a particular role to answer the phone, for testing purposes.

2. Run the script:

   python3 gpt-phone.py

Systemd Service

To run the GPT Phone as a systemd service, create a service file and enable it:

1. Enable and start the service:

   sudo cp phone.service /etc/systemd/systemd
   sudo systemctl --reload-daemon
   sudo systemctl enable phone.service
   sudo systemctl start phone.service
   loginctl enable-linger # this allows pulseaudio/pipewire to load upon boot

License

This project is licensed under the MIT License.