Automatic Note and Chord Recognition for HarmoniumMusic: A Deep Learning Approach
Surekha B. Puri, Dr.S. P. Mahajan · Journal of Critical Reviews · 2020
Music has played a significant role in the history of mankind. Understanding the theoretical fundamentals of music makes learning musical instruments easier. While many researchers have tried various techniques for music note recognition or acoustic chord recognition for the piano and other musical instruments, no method has been developed for the harmonium. A Convolutional way of Recurrent Neural Network known as(CRNN) based upon automated harmonium musical note identification approach is presented in this research by considering different audio features like pitch onset and offset times, signal energy and so on and their combinations. The audio samples used as input to the proposed system have been collected from a professional music player. These samples have been used to train the prediction model. In this approach, 900 harmonium audio samples comprising of various chord combinations are trained and tested by Convolutional Neural Network referred to as(CNN) and Recurrent Neural Network is also known as (RNN) with different sample combinations. CNN and RNN have a good rate of accuracy, but CRNN is the most accurate. The proposed system attains 94% accuracy. The resultant prediction of chords is then passed to the Lilypond library of Python to generate a music sheet that can be directly used by professional or novice musicians for composing music. The proposed method attains promising works on the self-created schema.