Los Angeles, CA
March 31, 2009 to April 2, 2009
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/CSIE.2009.253
The Regional style is one of the basic characteristic of Chinese folk songs. Because of the distinctive regional characteristics of Chinese folk songs, lots of folk songs lovers search for music by regional style. Therefore, geographical style automatic identification for folk songs is an important topic both for academic and industrial area. This paper studies geographical style automatic identification with different machine learning methods. An active feature selection method is proposed to improve the classification accuracy, and discover the most important feature for regional style classification. The experiments results show that SVM with active feature selection is an approximate best method. The classification accuracy of this method is 82.97%, and the features are reduced to 35 dimensions. Moreover, an improved combining multiple classifiers method can get the highest classification accuracy, that is 84.29%. Relative works show that our methods are also very efficient in other areas like genre classification.
Music information retrieval, music data mining, music style, music classification, feature selection
Yi Liu, Peng Wang, "Regional Style Automatic Identification for Chinese Folk Songs", CSIE, 2009, 2009 WRI World Congress on Computer Science and Information Engineering, CSIE, 2009 WRI World Congress on Computer Science and Information Engineering, CSIE 2009, pp. 5-9, doi:10.1109/CSIE.2009.253