International Computer Music Conference 2016 Papers Track

Full Program »

Do Structured Dichotomies Help in Music Genre Classification?

Dichotomy-based classification approaches are based on decomposing the class space of multiclass task into a set of binary-class ones. While they have been shown to perform well in classification tasks in other application domains, in this work we investigate whether they could also help improve genre classification in music, a core task in Mu- sic Information Retrieval. In addition to comparing some of the existing binary-class decomposition approaches, we propose and examine several new heuristics to build nested dichotomy trees. The intuition behind our heuristics is based on the observation that people find it easy to distinguish between certain classes and difficult between others. One of the proposed heuristics performs particularly well when compared to random selections from all possible balanced nested dichotomy trees. In our investigation, we use several base classifiers that are common in the literature and conduct series of empirical experiments on two music datasets that are publicly available for benchmark purposes. Additionally, we examine some issues related to the dichotomy-based approaches in genre classification and report the results of our investigation in this paper.

Author(s):

Tom Arjannikov    
Department of Computer Science
University of Victoria
Canada

John Zhang    
Department of Mathematics and Computer Science
University of Lethbridge
Canada

 

Powered by OpenConf®
Copyright©2002-2015 Zakon Group LLC