Müzik türlerinin otomatik sınıflandırılması için yükseltgeme (boosting) sınıflandırıcılarının kullanımı

Music genre classification is an important tool for music information retrieval systemsand has been finding important applications in various media platforms. Two importantproblems of the automatic music genre classification are feature extraction and classifier de-sign. There are recent works on these problems with promising future research directions.This thesis investigates discriminative boosting of classifiers to improve the automatic musicgenre classification performance. Two-class of classifiers, boosting of the Gaussian mixturemodel based classifiers and classifiers that are using the inter-genre similarity information,are proposed. Boosting is a technique that combines sequentially trained classifiers, wherein each new classifier a better modeling of hard-to-classify samples is done, and the over-all performance is boosted in the combined classifier. In this thesis a novel extension isproposed to the maximum-likelihood based training of the Gaussian mixtures to integrateGMM classifier into boosting architecture. Later, the boosting idea is modified to bettermodel the inter-genre similarity information over the mis-classified feature population. Oncethe inter-genre similarities are modeled, elimination of the inter-genre similarities reducesthe inter-genre confusion and improves the identification rates. Finally, an auto-clusteringscheme is build to determine similar music genre types for hierarchical classifier structure.Experimental results with promising identification improvements are provided.ii