While the core should be functional, there's still a lot to do.
- Fix TODOs and "Not implemented" in code.
- Improve API and modernize code
- Performance improvements (memory allocation!)
- Modernize model file format (Protobuf?)
- Prebuilds for common plattforms?
- Provide Website/API for hotword training (for non-technical users)
- Allow training with more than 3 samples
- Allow training with false positives
- Maybe implement speaker ID since we already do most of the work required for it?