Classification of musical patterns using variable duration Hidden Markov models

This paper presents a new extension to the variable duration Hidden Markov model, capable of classifying musical pattens that have been extracted from raw audio data, into a set predefined classes. Each musical pattern is converted into a sequence of music intervals by means of a fundamental frequency tracking procedure and it is subsequently given as input to a set of variable duration Hidden Markov models. Each of these models has been trained to recognize patterns of the respective predefined class. Classification is determined based on the highest recognition probability. This new type of variable duration Hidden Markov model provides increased classification accuracy because (a) it deals effectively with errors originating during the feature extraction stage and (b) it accounts for variations due to the expressive performance of instrument players. To demonstrate its effectiveness, the novel classification scheme has been employed in the context of Greek traditional music, to monophonic musical patterns of a popular instrument, the Greek Traditional clarinet. The classification results demonstrate that the new approach outperforms previous work based on conventional Hidden Markov models.