关键词: 音乐流派分类, 节拍特征, Mel频率倒谱系数（MFCC）, 小波分解, 支持向量机

Abstract: As the music beat features such as intensity, speed and duration which belong to semantic features characterize the different music genres, while the music beats are mostly generated by percussion component of the low-frequency part which can be extracted by using 6 layers wavelet decomposition of music signals. For genres which are not easy to be differentiated by beats features only, a method by combining the acoustic characteristics describing the frequency-domain energy envelope MFCC and beat features is proposed, and 8-order MFCC is used instead of the usual 12-order MFCC based on the music genres mechanism analysis. Music genres experiments simulation results evaluated on 8 genre classes show that the method based on combination of semantic and acoustic characteristics, the overall classification accuracy rate reaches 68.37%, while the increase of feature dimensions has little increase in the time-consuming.

Key words: music genre classification, beat features, Mel-Frequency Cepstral Coefficients（MFCC）, wavelet decomposition, Support Vector Machine（SVM）