I play a harmonica in my free time. One day, I was playing my harmonica and suddenly realized that the note that I played a while ago was wrong. Out of the blue, another idea popped in my mind. The idea was to build an AI to identify the note I play and tell me if I played the right note.

How does it work?

  • An audio file is recorded by the interface. This audio file contains the waveform for the note that is played by the user. To begin with, we trim the leading and trailing silence and then we break the audio signal into multiple overlapping parts to meet with the CNN's fixed input requirement
  • A spectrogram is plotted for each window and fed to the CNN to classify it. 
  • A majority vote of the CNN classifications is done in order to find an overall opinion

The full source code and details cannot be disclosed yet. Sorry 😞