Classification of Music by Genre Using Probabilistic Models and Deep Learning Models

Abstract

The digital shift in the distribution of music has presented the need for effective automated classification of large volumes of music into various categories. In this paper, automatic music genre classification is performed by first identifying and extracting representative aspects of a music piece. Subsequently, music features are tested for their significance in the task of genre classification using mutual information gain in order to make the feature vector compact, comprehensive and efficient. After several off-the-shelve classifiers were used, Support Vector Machines with a radial basis function kernel turned out to be the best performing model achieving an accuracy of 80.80 % in classifying music pieces into 1 of 10 genres in GTZAN.