Accoustic switch using mediapipe audio classification
The mediapipe framework provides an API for classification of audio (sound) samples in realtime. This can be used to classifiy the sounds of a microphone input signal (Try it).
The task is to do keyboard/mouse emulation based on a configured input sound/noise. The following tasks are necessary:
- Implement audio classification in python (see mediapipe audioclassifier)
- Add a gui to configure and assign a given sound class to a specific emulation
- Optional
- Retrain the used classification ML model to recognize other sounds (e.g. self-recorded whistle, plop sounds) using transfer learning (see tensorflow tutorial for audio classification).
- Optimize model and reduce size for Raspberry Pi and smaller devices using Tensorflow lite model architecture.