Automatic classification of musical genres using inter-genre similarity

Summary

The paper addresses feature extraction and classifier design for automatic music genre classification. It proposes two classifiers based on inter-genre similarity (IGS), modeled using difficult-to-classify samples in the genre feature space. The authors also investigate dynamic timbral texture features. They report that removing samples in the modeled inter-genre similarity class reduced confusion between genres and that the proposed classifiers achieved better classification rates than the existing methods they evaluated. The abstract does not specify the evaluation dataset or establish performance beyond those experiments.