Capturing Genre Characteristics of Music Through Feature-Value Analysis

In this paper, we present our initial investigations on how to capture genre characteristics through feature-value pairs in large music datasets. In our approach, we propose an adaptation of association analysis, a computational technique to explore the inherent relationships among data objects in a problem domain, to detect and extract acoustical features of music genres and use the resultant relationships among them to classify new music pieces. In essence, for each individual genre, we obtain a set of characteristic features and their values that represent it. The effectiveness of our approach is demonstrated through empirical experiments on a number of music datasets. The results are presented and discussed. Various related issues are examined.