Boosting classifiers for music genre classification

Summary

The paper investigates discriminative boosting for automatic music genre classification. It presents two classifier structures: one integrates Gaussian mixture model (GMM) classifiers into a boosting framework; the other uses inter-genre similarity information to model misclassified features and reduce confusion between genres. The authors also describe a hierarchical auto-clustering scheme for the similarity-based approach. The abstract reports experimental classification improvements but does not provide performance figures; this summary does not imply generalization beyond the reported evaluation. Read the paper.