Content-based music genre classification using timbral feature vectors and support vector machine

Abstract

This paper proposes a novel content-based music genre classification method using timbral feature vectors and support vector machine (SVM). The timbral feature vectors used in the proposed method consist of both the long-term and the short-term features which can represent the time-varying behavior of music. These features are mel-frequency cepstral coefficient (MFCC) plus log energy with different frame length. The timbral feature vectors will be applied to train an optimized non-linear decision rule for music genre classifier via SVM. This paper selects nine kinds of different music, including classical, jazz, dance, lullaby, country, Bossa Nova, piano, blue note, and hip-hop, for performance evaluation. Experimental results show that the proposed method can achieve the average accuracy rate of 86% for the nice music genres classification.