A Music genre classification using music features and neural networks

Abstract

Genre is a conventional way to classify music. This paper aims to implement an algorithm to classify music files into four different genre, which are classic, rock, pop, and dangdut. Dangdut is Indian influenced Indonesian popular music. An automatic music genre classification system will help music collectors to classify their music collection. The genre classification algorithm is divided into music feature extractor and classifier. The feature extraction of the system is constructed by timbre feature extraction and rhythm feature extraction. The timbre feature extraction is done by mel frequency cepstral coefficients (MFCC). The rhythm feature is constructed by full wave rectification, low pass filtering, down sampling, mean removal, and autocorrelation. The classifier is formed by back propagation neural network (BPNN). By using this algorithm, the average of accuracy is 80%, that is of 100% accuracy for classic, 70% accuracy for rock, 90% accuracy for pop, and 60% accuracy for dangdut.