# Wake-Word Detection Service

Wake-word detection service using openWakeWord for detecting "Hey Atlas".

## Features

- Real-time wake-word detection using openWakeWord
- WebSocket events for detection notifications
- HTTP API for control (start/stop)
- Low-latency audio processing
- Configurable threshold

## Installation

```bash
# Install system dependencies (Ubuntu/Debian)
sudo apt-get install portaudio19-dev python3-pyaudio

# Install Python dependencies
pip install -r requirements.txt
```

## Usage

### Standalone Service

```bash
# Run as HTTP/WebSocket server
python3 -m wake-word.server

# Or use uvicorn directly
uvicorn wake-word.server:app --host 0.0.0.0 --port 8002
```

### Python API

```python
from wake_word.detector import WakeWordDetector

def on_detection():
    print("Wake-word detected!")

detector = WakeWordDetector(
    wake_word="hey atlas",
    threshold=0.5,
    on_detection=on_detection
)

detector.start()
# ... do other work ...
detector.stop()
```

## API Endpoints

### HTTP

- `GET /health` - Health check
- `GET /status` - Get detection status
- `POST /start` - Start wake-word detection
- `POST /stop` - Stop wake-word detection

### WebSocket

- `WS /events` - Receive wake-word detection events

**WebSocket Message Format:**
```json
{
  "type": "wake_word_detected",
  "wake_word": "hey atlas",
  "timestamp": 1234.56
}
```

## Configuration

- **Wake-word**: "hey atlas" (default)
- **Sample Rate**: 16000 Hz
- **Threshold**: 0.5 (confidence threshold)
- **Chunk Size**: 1280 samples

## Integration

The wake-word service emits events that trigger:
1. ASR service to start capturing audio
2. LLM processing pipeline
3. TTS response

## Testing

```bash
# Test detector directly
python3 -m wake-word.detector

# Test HTTP server
curl http://localhost:8002/health
curl -X POST http://localhost:8002/start
curl -X POST http://localhost:8002/stop
```

## Notes

- Requires microphone access
- Uses openWakeWord (Apache 2.0 license)
- For custom wake-words, need to train a model
- Default model may need fine-tuning for "Hey Atlas"