结合节拍语义和MFCC声学特征的音乐流派分类  

Combining beat semantic features and MFCC acoustic features for music genre classification

作　　者：庄严[1] 于凤芹[1] 

机构地区：[1]江南大学物联网工程学院,江苏无锡214122

出　　处：《计算机工程与应用》2015年第3期197-201,共5页Computer Engineering and Applications

基　　金：国家自然科学基金(No.61075008)

摘　　要：由于音乐节拍的强度、快慢、持续时间等是反映音乐不同流派风格的重要语义特征,而音乐节拍多属于由打击乐器所产生的低频部分,为此利用小波变换对音乐信号进行6层分解来提取低频节拍特征;针对节拍特征差异不明显的音乐流派,提出用描述频域能量包络的MFCC声学特征与节拍特征结合,并用基于音乐流派机理分析的8阶MFCC代替常用的12阶MFCC。对8类音乐流派实验仿真结果表明,基于语义特征和声学特征结合的方法,总体分类准确率可达68.37%,同时特征维数增加对分类时间影响很小。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.

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 

分 类 号：TP37[自动化与计算机技术—计算机系统结构]