Automatic Music Transcription for Monophonic Piano Music via Image Recognition

Music transcription has been a longtime challenging task even for human. It takes a significant amount of time and effort for an experienced musician to listen to a song or music and transcribe it into music sheets. Automatic Music Transcription (AMT) automates the process of transcribing musics and plays an important role in music information retrieval(MIR). Even though the research for AMT is still in infancy, the results so far have been proved to be very educational to both the areas of Machine Learning and Music Composition. In this project, we tried to tackle the problem of music transcription using a new approach proposed by [1] that transforms the music note detection problem into an image recognition problem using Convolutional Neural Network (CNN). We gathered our training and testing data from the MAPS database [2] which contains recordings and MIDI files of isolated piano notes and built upon the existing CNN image recognition algorithm with Tensorflow for the music transcription problem.