DiaOfflineAI: User-Friendly Offline Voice Assistant for Raspberry Pi

An entirely offline voice assistant that runs on Raspberry Pi to detect a custom wake-word, transcribe speech, generate responses, and speak them back in a custom TTS voice. DiaOfflineAI focuses on user-friendliness, allowing non-technical users to easily install, customize, and use a powerful voice assistant without cloud dependencies.

Hardware Requirements

• Raspberry Pi 5 (16 GB)
• 64 GB A2 microSD (OS, swap, logs)
• 512 GB NVMe SSD on M.2 HAT+ (large models, RAG store)
• Coral USB Edge TPU + Hailo-8L AI Kit (dual accelerator)
• ReSpeaker 4-Mic Array HAT (far-field voice capture)
• HiFiBerry DAC+ HAT (speaker output)
• Official 27 W USB-C PSU, active-cooler case, GPIO spacers

Software Components

1. Wake-Word Engine

   • Porcupine SDK with custom "Hey Dia" model (.ppn)

2. Offline ASR

   • Vosk small-en US model for 16 kHz microphone input

3. Local LLM (optional / future)

   • Integration with llama.cpp or similar for on-device chat

4. TTS

   • Dia-TTS ("dia-expressive") to synthesize responses as WAV
   • Custom voice creator and profile manager

5. Audio I/O

   • PyAudio for capture/playback, ALSA config tuned for ReSpeaker & DAC
   • Easy Bluetooth headset integration

6. AI Acceleration

   • PyCoral / TensorFlow Lite for TPU, plus Hailo SDK integration
   • Automatic performance optimization

7. Persistence & RAG

   • FAISS+SQLite on NVMe for retrieval of user data or documents
   • Simple knowledge pack installation

8. User-Friendly Interfaces

   • Visual feedback for speech recognition
   • Control Center GUI for easy management
   • Personality customization

Project Structure

dia-assistant/
├── config/             # Configuration files for components
│   ├── alsa/           # ALSA configurations for audio hardware
│   ├── systemd/        # Systemd service unit
│   └── dia.yaml        # Main configuration
├── docs/               # Documentation
├── logs/               # Log files (symlinked to /var/log/dia)
├── models/             # Model storage directory
│   ├── asr/            # Vosk ASR models
│   ├── llm/            # LLM models
│   ├── tts/            # TTS voice models
│   └── wake/           # Wake word models
├── scripts/            # Utility scripts
│   ├── setup_dia.sh    # Main installation script
│   ├── model_download.sh # For downloading models
│   ├── dia-voice.sh    # Voice profile manager
│   ├── dia-custom-voice.sh # Custom voice creator
│   ├── dia-knowledge.sh # Knowledge pack manager
│   ├── dia-optimize.sh # Performance optimization
│   ├── dia-personality.sh # Personality customizer
│   ├── dia-bluetooth.sh # Bluetooth device manager
│   ├── dia-visual-test.py # Speech recognition visualizer
│   ├── dia-easy.sh     # Easy launcher for all functions
│   ├── dia-control-center.py # Graphical user interface
│   └── install-control-center.sh # GUI installer
├── src/                # Source code
│   ├── asr/            # ASR module
│   ├── llm/            # LLM/response generation
│   ├── rag/            # RAG components
│   ├── tts/            # TTS module
│   ├── wake/           # Wake word detection
│   ├── audio/          # Audio handling
│   ├── utils/          # Utility functions
│   └── dia_assistant.py # Main application entry point
├── Dockerfile          # For containerization
├── docker-compose.yml  # Docker configuration
└── requirements.txt    # Python dependencies

Installation

1. Clone this repository:

   git clone https://github.com/your-org/DiaOfflineAI.git
   cd DiaOfflineAI

2. Run the easy setup script:

   ./scripts/easy-setup.sh

The setup script will:

• Create a Python virtual environment
• Install required dependencies
• Download necessary models
• Configure hardware components
• Set up systemd service
• Create desktop shortcuts
• Install the Dia Control Center

Usage

Once installed, DiaOfflineAI will automatically start on boot. The easiest way to manage it is through the graphical Control Center:

# Launch the graphical control center
sudo dia-control

Alternatively, you can use command-line tools:

# Launch the easy management menu
dia-easy

# Or use the traditional service commands
sudo systemctl start dia.service
sudo systemctl status dia.service
sudo systemctl stop dia.service

User-Friendly Features

Visual Control Center

The Dia Control Center provides an intuitive graphical interface with:

• Simple start/stop/restart buttons
• Status indicator showing if Dia is running
• Quick access to all tools organized by category
• Clean, modern interface that's easy to navigate

Voice & Personality

# Voice profile manager
dia-voice

# Custom voice creator
dia-custom-voice

# Personality customizer
dia-personality

Knowledge Management

# Install specialized knowledge packs
dia-knowledge

Hardware & Performance

# Set up Bluetooth devices
dia-bluetooth

# Test speech recognition visually
dia-visual

# Optimize performance
dia-optimize

Traditional Customization

Adding Custom Wake Words

Place your .ppn Porcupine wake word model files in the models/wake/ directory and update the configuration in config/dia.yaml.

Replacing ASR/TTS Models

1. Download new models to the appropriate directory:

   ./scripts/model_download.sh --asr large-model
   ./scripts/model_download.sh --tts new-voice

2. Update the model paths in config/dia.yaml

RAG Updates via USB

1. Prepare your documents on a USB drive with a specific format
2. Insert the USB drive into the Raspberry Pi
3. Run the RAG update script:

   ./scripts/update_rag.sh --source /media/usb

Troubleshooting

Use the Control Center to quickly access logs and status information, or check logs at /var/log/dia/ using:

journalctl -u dia.service

Common issues:

• Audio device not found: Use dia-bluetooth to reconfigure audio devices
• Wake word not detected: Run dia-visual to test microphone input visually
• High CPU usage: Use dia-optimize to tune performance

License

MIT License