Automatic musical genre classification of OPM songs

Abstract

One of the factors that is known to affect song choices is an individual’s musical genre preference. Accurate classification of a song’s genre will facilitate in identifying which market to target in the release of a potential hit song. We extract 56 single-valued musical features from 380 Original Pilipino Music (OPM) songs (120 ballad, 120 alternative rock and 140 rock songs) from 2004 to 2006. Based on the effect size measure of a variable’s ability to discriminate between hit and non-hit songs, the 20 highest-ranked features are fed to a classifier tasked to determine a song’s musical genre. We show that a system developed for hit song prediction is also at the same time capable of musical genre classification. Near perfect genre classification accuracies are attained with a trained feed-forward neural network (NN) and linear discriminant analysis (LDA) but only around 77% accuracy using Classification and Regression Tree (CART). All of the classifiers employed are well above the suggested acceptable classifier requirement of 41.7%. In addition, parameters that are measures of a song’s melodic interval contributed significantly to the success of the genre classification, regardless of the method used.