Music Classification Using Fourier Transform and Support Vector Machines

Abstract

Information retrieval from music is an active research area in computer science. In this paper, we perform a music classification by genre using the subset of the characteristics of the music signal. Features based on magnitude, pitch, and tempo have been found to be informative for classifying musical pieces by genre. We group the features into these categories. These features are calculated from the Fourier transform’s magnitude spectrum. By analyzing the data and exploring it, we develop knowledge about features that can be used for classification, and finally using an information ranking classifier to select the best feature. Finally, Support Vector Machines had the best performance with an accuracy of 81.85% when classifying Spotify music into 20 genres.