Music Genre Classification Using Temporal Information and Support Vector Machine

This paper proposes a novel content-based music genre classification method using temporal information and support vector machine. By processing of texture window statistics and delta computation, temporal information is incorporated to capture the time-varying behavior of music. 7 different temporal evolution descriptors applied in texture window are introduced in this paper, i.e., mean, standard deviation, maximum, minimum, temporal centroid, temporal skewness and temporal kurtosis. 4 different ways of representing a music clip are proposed and compared. The experimental results show that temporal information modeled by delta computation and texture window statistics is very effective and efficient in music genre classification. This paper uses a dataset consisting of 1000 30-sec music clips equally divided into 10 music genres, namely classical, blues, jazz, country, rock, metal, reggae, hiphop, disco and pop. The best classification accuracy achieved is 78.6% for this 10-class music genre classification.