A Consideration on Automatic Music Genre Classification Using Rough Set Theory
Automatic music genre classification is important for music information retrieval and automatic music selection. Herein the applicability of rough set theory to music genre classification is examined. Rough set theory is a mathematical theory to approximate concepts, and is widely used in the field of data mining. In rough set theory applications, data of objects are generally expressed by a combination of attributes and their values in a so-called decision table. Decision rules to determine the class of an object are extracted based on the attributes of the object. In music genre classification, the object is a piece of music, and the class of the object is the genre. Herein ten acoustical parameters of a musical piece are proposed as attributes of a decision table; seven parameters are related to the long-term spectrum of the piece, and three are associated with the waveform. Based on these parameters, decision rules for music genre classification were extracted using 40 pieces based on the lower approximation. These rules were applied to 80 additional pieces to examine the reliability of the extracted rules. Approximately 80% of the pieces were correctly classified into two major genres: Classic/Jazz or Rock/Metal. Moreover, the results suggest that parameters related to the waveform of a piece are significant for automatic music genre classification.
Kenji Ozawa Yuka Masaki Tomohiko Ise
University of Yamanashi,Kofu,400-8511,Japan Alpine Electronics,Inc.,Iwaki,970–1192,Japan
国际会议
The 10th Western Pacific Acoustics Conference(第十届西太平洋声学会议WESPAC X)
北京
英文
1-8
2009-09-21（万方平台首次上网日期，不代表论文的发表时间）