Phase 1: Wake Word Detection
Goal
Get "Hey LARS" wake word working on a device.
Tasks
1. Train Custom Wake Word
- [ ] Access openWakeWord Google Colab
- [ ] Record/generate training samples for "Hey LARS"
- [ ] Train model (~1 hour)
- [ ] Export .tflite file
2. Test Detection
- [ ] Install openWakeWord on local-ai server
- [ ] Load custom model
- [ ] Test with microphone input
- [ ] Measure accuracy and false positives
3. Create Listener Service
- [ ] Python script that runs continuously
- [ ] Listens for wake word
- [ ] On detection: trigger next phase
- [ ] Systemd service for auto-start
Deployment Options
- local-ai server - Already has hardware
- Raspberry Pi - Dedicated always-on device
- Tablet/Phone app - PWA with microphone access
- Web browser - For testing
Success Criteria
- "Hey LARS" detected reliably
- < 1 second detection latency
- Low false positive rate
- Runs 24/7 without issues