Automatic Segmentation and Comparative Study of Motives in Eleven Folk Song Collections using Self-Organizing Maps and Multidimensional Mapping

Abstract

A data-based system for automatic segmentation of large folk song corpora is described in this article. The algorithm is based on a self-organizing map that learns the most typical motive contours. Using this system, the typical motive collections of 11 cultures in Eurasia were determined. The analysis of the overlaps between the cultures allowed us to draw a graph of connections, which shows two main distinct groups, according to the geographical distribution. These groups are connected by the cultures of the Carpathian Basin, which in itself assures the unbroken structure of the system of connections. The mapping of the motive contours into points of an appropriate three-dimensional space opened the possibility to analyse the musical structures of the typical motives in different cultures. Based on the segmentation algorithm, we also defined a melody similarity measure, determining local similarities between the closest motive contours.