Genre Classification in Music using Convolutional Neural Networks

With the advancement of technology and computational power, crafting a chart-topping song has become more effortless than before, achievable from the convenience of our residences with just a computer at hand. This has led to the emergence of vast arrays of catalogs of music, containing a variety of genres and styles from different music makers with different ethnicities and backgrounds, resulting in a large database that clogs most music streaming platforms with little automated categorization. Based on the GTZAN audio dataset, this paper revisits the use of Convolution Neural Networks (CNN) for classifying different types of music genres. Using Mel-frequency cepstral coefficients (MFCC) features, the CNN model achieved an accuracy of 85%. As a result of the careful design of the CNN model, it is on par with many latest and greatest CNN frameworks.