Automatic Music Transcription

Automatic Music Transcription is one of the most interesting topics that I have studied from de signal processing point of view. In short, this is about automatically transcribing audio files (e.g. wav file) into music notation formats, such as MIDI or musical score.


During the course Audio Analysis and Processing (Spring 2011) taught by Professor Colby Leider at University of Miami, I made a simple automatic music transcription system based on comb filters for polyphonic audio. You can download the paper and the code.

Some examples:


As you can perceive, the system performs quite well for multi-pitch estimation. However, onset detection does not work properly sometimes, which is highly noticeable in the example Beatles-Something.