The Feature Selection of Regional Style Classification of Chinese Folk Songs

刘怡,蔚磊,刘子利,王鹏

DOI: https://doi.org/10.3321/j.issn:0372-2112.2008.z1.027

2008-01-01

Abstract:Automatic music style classification is an important topic in the area of music information retrieval.In this paper,we present a study on automatic classification of Chinese folk songs by regional style which mainly focuses on performance of different feature selection method.We did music style classification experiments on 1392 original Chinese folk songs which are collected from 10 different regions.The experiment results show that support vector machine classification performance has a certain advantage among different classifiers without feature selection.Simultaneously,support vector machine get the highest classification accuracy of 83% with active feature selection method,the feature vector dimensions are reduced from 74 to 35 using active feature selection feature selection.Therefore,the selected feature set is more convenient for music analysis.