Wavelet-filtering of symbolic music representations for folk tune segmentation and classification

The aim of this study is to evaluate a machine-learning methodin which symbolic representations of folk songs are segmentedand classified into tune families with Haar-wavelet filtering.The method is compared with previously proposed Gestalt basedmethod. Melodies are represented as discrete symbolicpitch-time signals. We apply the continuous wavelet transform(CWT) with the Haar wavelet at specific scales, obtaining filteredversions of melodies emphasizing their information at particulartime-scales. We use the filtered signal for representationand segmentation, using the wavelet coefficients’ local maximato indicate local boundaries and classify segments by means ofk-nearest neighbours based on standard vector-metrics (Euclidean,cityblock), and compare the results to a Gestalt-based segmentationmethod and metrics applied directly to the pitch signal.We found that the wavelet based segmentation and wavelet filteringof the pitch signal lead to better classification accuracyin cross-validated evaluation when the time-scale and other parametersare optimized