Content-Based Feature Selection for Music Genre Classification

Abstract

The most important aspect that one should consider in a content-based analysis study is the feature that represents the information. In music analysis one should know the details of the music contents that can be used to differentiate the songs. The selection of features to represent each music genre is an important step to identify, label, and classify the songs according to the genres. This research investigates, analyzes, and select timbre, rhythm, and pitch-based features to classify music genres. The features that were extracted from the songs consist the singer’s voice, the instruments and the melody. The feature selection process focuses on the supervised and unsupervised methods with the reason to select significant generalized and specialized music features. Besides the selection process, two modules of Negative Selection Algorithm; censoring and monitoring are highlighted as well in this work. We then proposed the Modified AIS-based classification algorithm to solve the music genre classification problem. The results from our experiments demonstrate that the features selection process contributes to the proposed modified AIS-based music genre classification performs significantly in classifying the music genres.