Description

In phase one of the wakeword project, we delivered prototypes for data collection, and training of sparse wakeword models. In phase 2, we focus on usability of the wakeword engine, especially with constrained resources.

Deliverables

• tflite model
• Use Wake Word Data Prep models to test TF lite compression optimization
• Precise TFlite with compression repo
• Precise Rust
  • Runner
  • MFCC in rust
• Precise Android

DoD (definition of done)

• Unit testing
• End2End test: train and run inference
• Models must pass production quality control: 1 week: wake up every time, ~2-3 false wake ups (future goal: 1 false wake up per week!)
• Benchmark performance: CPU% raspi4 (phone if possible), loss of quality of model