🗣️ Spill The GPT

An interactive AI gossip installation created for Dutch Design Week 2025. This system transforms personal stories into anonymized gossip using AI, creating a chain of whispered secrets through an old telephone handset.

🎯 Overview

Spill The GPT is an interactive art installation that uses AI to collect, anonymize, and share gossip. Users pick up an old telephone, hear gossip from a previous participant, then share their own story. The system uses OpenAI's Whisper for speech recognition, GPT-4 for text transformation, and ElevenLabs for text-to-speech, creating an endless chain of anonymous stories.

Two Modes of Operation

1. Web Interface (app.py) - Browser-based gossip collection for development and testing
2. Phone System (enhanced_phone_system.py) - Raspberry Pi installation with physical telephone handset

🎭 How It Works

Enhanced User Flow:

1. 📞 Phone Pickup - User lifts the handset
2. 🔊 Welcome Message - "Welcome to the gossip system. First you will hear gossip from a previous user."
3. 🎧 Previous Gossip - Plays a random gossip from the database
4. 🔄 Transition Message - "Now it's your turn. Tell your gossip and hang up when you're done."
5. 🎙️ Recording - User speaks their gossip
6. 📞 Phone Hangup - Stops recording and begins processing
7. 🤖 AI Processing Chain:
   • Whisper transcribes the audio to text
   • GPT-4 transforms it into anonymized gossip
   • ElevenLabs converts it back to speech (Dutch voice)
8. 💾 Storage - Saved in SQLite database for future playback
9. 🔁 Loop - Next user hears this gossip

🛠️ Hardware Requirements

For Raspberry Pi Installation:

• Raspberry Pi (3B+ or newer recommended)
• Old telephone handset with hookswitch
• Magnetic reed switch or button sensor (hookswitch detection)
• USB microphone or Raspberry Pi audio HAT
• Speaker or amplified audio output
• Jumper wires and breadboard
• Power supply for Raspberry Pi

GPIO Wiring:

• Hookswitch Sensor: GPIO 18 (configurable)
• Ground: GND pin
• Power: 3.3V (if needed)

📦 Installation

Prerequisites

1. System Dependencies (Raspberry Pi)

sudo apt update
sudo apt install -y portaudio19-dev python3-pyaudio alsa-utils pulseaudio

2. Python Environment

# Create virtual environment
python3 -m venv env

# Activate environment
# On Linux/Mac:
source env/bin/activate
# On Windows:
env\Scripts\activate

# Install dependencies
pip install -r requirements_enhanced.txt

3. API Keys Setup

Create a .env file in the project root:

OPENAI_API_KEY=your_openai_api_key_here
ELEVENLABS_API_KEY=your_elevenlabs_api_key_here

Or export as environment variables:

export OPENAI_API_KEY="your_openai_api_key"
export ELEVENLABS_API_KEY="your_elevenlabs_api_key"

4. Generate Welcome Audio Messages

python create_audio_messages.py

This creates:

• welcome.mp3 - Initial greeting
• transition.mp3 - "Now it's your turn" message

🚀 Running the System

Web Interface Mode (Development/Testing)

python app.py

Access at http://localhost:5000

Features:

• Record audio through browser
• Transcribe and process gossip
• View gossip loop of all collected audio

Phone System Mode (Raspberry Pi)

python enhanced_phone_system.py

The system will wait for the phone to be picked up and automatically handle the interaction flow.

📂 Project Structure

Spill_The_GPT/
├── app.py                          # Flask web application
├── enhanced_phone_system.py        # Main phone system for Raspberry Pi
├── gossip_database.py              # SQLite database management
├── el.py                           # ElevenLabs TTS integration
├── create_audio_messages.py        # Generate welcome/transition audio
├── requirements.txt                # Web app dependencies
├── requirements_enhanced.txt       # Enhanced system dependencies
├── templates/                      # HTML templates
│   ├── index.html                 # Main web interface
│   └── gossip_loop.html           # Continuous gossip playback
├── Raspberry_Pi_setup/            # Raspberry Pi specific files
│   ├── phone_system.py            # Main phone system for Raspberry Pi
│   ├── gossip_database.py         # Database handler
│   ├── el.py                      # ElevenLabs TTS handler
│   ├── create_welcome.py          # Welcome audio generator
│   ├── requirements_phone.txt     # Pi dependencies
│   ├── PHONE_SETUP.md            # Detailed setup guide
│   ├── CLEAN_PI_SETUP.md         # Clean installation guide
│   ├── setup_clean_pi.sh         # Automated installation script
│   ├── setup_kiosk_mode.sh       # Auto-start configuration script
│   ├── test_audio_devices.py     # Audio device diagnostics
│   ├── test_microphone.py        # Microphone testing utility
│   └── gpio_test.py              # GPIO testing utility
├── audio/                         # Generated gossip audio files
├── gossip.db                      # SQLite database (auto-created)
├── welcome.mp3                    # Welcome message audio
└── transition.mp3                 # Transition message audio

🔧 Configuration

Phone System Settings

Edit enhanced_phone_system.py:

HORN_BUTTON_PIN = 18        # GPIO pin for hookswitch
AUDIO_RATE = 44100          # Recording sample rate
AUDIO_CHANNELS = 1          # Mono recording
AUDIO_CHUNK = 1024          # Audio buffer size

Audio Device Configuration

Test and configure audio devices:

# List audio devices
aplay -l     # Playback devices
arecord -l   # Recording devices

# Test microphone
arecord -D plughw:1,0 -d 5 test.wav
aplay test.wav

# Adjust audio levels
alsamixer

Voice Customization

Edit el.py to change the ElevenLabs voice:

voice_id="ANHrhmaFeVN0QJaa0PhL"  # Currently: Petra (Flemish)
model_id="eleven_turbo_v2_5"      # Turbo model for low latency

🗄️ Database Structure

SQLite database (gossip.db) automatically created with:

CREATE TABLE gossip (
    id INTEGER PRIMARY KEY AUTOINCREMENT,
    file_path TEXT NOT NULL,
    original_text TEXT,              -- Raw transcription
    gossip_text TEXT,                -- GPT-4 anonymized version
    timestamp DATETIME DEFAULT CURRENT_TIMESTAMP,
    duration_seconds REAL,
    file_size_bytes INTEGER,
    is_active BOOLEAN DEFAULT 1
)

Database Features:

• Automatic Creation: No manual setup required
• Random Selection: Fair distribution of previous gossip
• Persistent Storage: Survives system restarts
• File Cleanup: Automatically handles missing audio files
• Metadata Tracking: Records timestamps, file sizes, and durations

🎨 Kiosk Mode Setup (Auto-Start on Boot)

Quick Setup

Run the automated script:

cd Raspberry_Pi_setup
chmod +x setup_kiosk_mode.sh
./setup_kiosk_mode.sh

Manual Systemd Service Setup

Create service file:

sudo nano /etc/systemd/system/phone-gossip.service

Add configuration:

[Unit]
Description=Phone Gossip System
After=network.target sound.service

[Service]
Type=simple
User=pi
WorkingDirectory=/home/pi/Spill_The_GPT
Environment=PATH=/home/pi/Spill_The_GPT/env/bin
ExecStart=/home/pi/Spill_The_GPT/env/bin/python enhanced_phone_system.py
Restart=always
RestartSec=10

[Install]
WantedBy=multi-user.target

Enable and start:

sudo systemctl daemon-reload
sudo systemctl enable phone-gossip.service
sudo systemctl start phone-gossip.service

🧪 Testing & Troubleshooting

Audio Device Testing

cd Raspberry_Pi_setup
python test_audio_devices.py    # List and test all audio devices
python test_microphone.py        # Record and playback test

GPIO Testing

cd Raspberry_Pi_setup
python gpio_test.py              # Test hookswitch button

Common Issues

Audio Not Working

• Check device permissions: sudo usermod -a -G audio $USER
• Verify audio devices with aplay -l and arecord -l
• Test with speaker-test and arecord
• Adjust levels in alsamixer

GPIO Issues

• Verify wiring connections
• Check pin permissions
• Test with gpio readall (install wiringpi)
• Ensure correct pin numbering (BCM vs BOARD)

Database Issues

• Check file permissions in project directory
• Verify database with: sqlite3 gossip.db ".tables"
• Database auto-creates on first run

Performance Issues

• Use Whisper "tiny" model for faster transcription
• Monitor CPU temperature: vcgencmd measure_temp
• Consider using lighter weight models
• Reduce audio quality if needed

🌐 Web Interface Features

Access the web interface at http://localhost:5000

Main Interface (/)

• Record audio directly in browser
• Live transcription
• GPT-4 anonymization
• Text-to-speech playback

Gossip Loop (/gossip-loop)

• Continuous playback of all gossip
• Auto-advancing carousel
• Background installation mode

🔐 Security & Privacy

• Anonymization: GPT-4 removes personally identifiable information
• No User Tracking: System doesn't store user identities
• Local Storage: All data stored locally in SQLite
• Audio Cleanup: Temporary recordings automatically deleted
• API Keys: Store in .env file (not in version control)

🎯 Use Cases

1. Art Installations: Interactive gallery pieces
2. Events: Dutch Design Week, festivals, exhibitions
3. Research: Social interaction and AI studies
4. Education: Demonstrating AI pipeline (STT → LLM → TTS)

📝 Development Notes

First User Experience

• When no previous gossip exists, system handles gracefully
• First gossip becomes available for second user
• Database grows organically with each interaction

Whisper Model Options

• tiny: Fastest, lowest accuracy (recommended for Pi)
• base: Good balance
• small: Better accuracy, slower
• medium/large: Best accuracy, very slow on Pi

ElevenLabs Configuration

• Currently uses Dutch/Flemish voice (Petra)
• Turbo model for low latency
• Configurable speed, stability, and similarity

🤝 Contributing

This project was created for Dutch Design Week 2025. Feel free to fork and adapt for your own installations.

📄 License

Created for Dutch Design Week 2025

🙏 Acknowledgments

• OpenAI: Whisper and GPT-4 models
• ElevenLabs: Text-to-speech voice synthesis
• Dutch Design Week: Exhibition platform

📞 Support

For issues or questions, please refer to the setup guides:

• Raspberry_Pi_setup/PHONE_SETUP.md - Phone system guide
• Raspberry_Pi_setup/CLEAN_PI_SETUP.md - Clean installation guide
• ENHANCED_SYSTEM.md - Enhanced features documentation

---

Project: Spill The GPT
Event: Dutch Design Week 2025
Type: Interactive AI Art Installation