A computational analysis of Turkish makam music based on a probabilistic characterization of segmented phrases

Abstract

This study targets automatic analysis of Turkish makam music pieces on the phrase level. While makam is most simply defined as an organization of melodic phrases, there has been very little effort to computationally study melodic structure in makam music pieces. In this work, we propose an automatic analysis algorithm that takes as input symbolic data in the form of machine-readable scores that are segmented into phrases. Using a measure of makam membership for phrases, our method outputs for each phrase the most likely makam the phrase comes from. The proposed makam membership definition is based on Bayesian classification and the algorithm is specifically designed to process the data with overlapping classes. The proposed analysis system is trained and tested on a large data set of phrases obtained by transferring phrase boundaries manually written by experts of makam music on printed scores, to machine-readable data. For the task of classifying all phrases, or only the beginning phrases to come from the main makam of the piece, the corresponding F-measures are .52 and .60 respectively.