Musical Genre Classification using Gaussian Mixture Models

Abstract

Digital music is one of the most important data types, distributed by the Internet. Automatic musical genre classification is very useful for music indexing and retrieval. A method to recognize the genre of music audio is considered. In this paper, the input music is represented with DWT (Discrete Wavelet Transform) coefficients and classifying the extracted features is performed using Gaussian Mixture Models (GMM). Using GMM the optimal class boundaries between four groups of genre namely, pop, classic, rock and jazz are obtained. The feature vector from feature extraction step uses wavelet coefficients by hierarchical decomposition as it is easy to implement as well as it can reduce the computation time and resources required. Given that GMM is a robust approach that could obtain very good performance and a solution based on it is powerful, the classification is mainly composed of GMM classifiers. The experimental results indicate that the proposed approach offer encouraging results.