Music Genre Classification using Convolutional Neural Networks

Abstract

Music plays a vital role in one’s life, the type of music and the labels created by humans to classify music. There are different characteristics associated with the rhythm and instruments used to create music. We get huge collections of music available online. But at present many companies depend on human work to classify the genre. Our project was mainly to make an automatic system for classification models for music genres. This will replace humans in future and can get good accuracy which will be an added advantage for the music retrieval systems. A complete set of seven features namely MFCC vector of 20 numerical values, chroma frequencies, spectral rolloff, spectral centroid, zero-crossing rate, spectral bandwidth and root mean square were used for obtaining features we are using GTZAN genre dataset. The project is mainly developed using CNN to classify the musical genres. At last we get the output stating what genre the music belongs to at the accuracy of 74%.