Using Digraphs and a Second-Order Markovian Model for Rhythm Classification

The constant increase of online music data has required reliable and faster tools for retrieval and classification of music content. In this scenario, music genres provide interesting descriptors, since they have been used for years to organize music collections and can summarize common patterns in music pieces. In this paper we extend a previous work by considering digraphs and a second-order Markov chain to model rhythmic patterns. Second-order transition probability matrices are obtained, reflecting the temporal sequence of rhythmic notation events. Additional features are also incorporated, complementing the creation of an effective framework for automatic classification of music genres. Feature extraction is performed by principal component analysis and linear discriminant analysis techniques, whereas the Bayesian classifier is used for supervised classification. We compare the obtained results with those obtained by using a previous approach, where a first-order Markov chain had been used. Quantitative results obtained by the kappa coefficient corroborate the viability and superior performance of the proposed methodology. We also present a complex network of the studied music genres.