A simple-cycles weighted kernel based on harmony structure for similarity retrieval

Abstract

This paper introduces a novel methodology for music similarity retrieval based on chord progressions. From each chord progression, a directed labeled graph containing the interval transitions is extracted. This graph will be used as input for a graph comparison method based on simple cycles – cycles where the only repeated nodes are the first and the last one. In music, simple cycles represent the repetitive sub-structures of, e.g., modern pop/rock music. By means of a kernel function [10] whose feature space is spanned by these simple cycles, we obtain a kernel matrix (similarity matrix) which can then be used in music similarity retrieval tasks. The resulting algorithm has a time complexity ofO(n+m(c+1)), where n is the number of vertices, m is the number of edges, and c is the number of simple cycles. The performance of our method is tested on both an idiom retrieval task, and a cover song retrieval task. Empirical results show the improved accuracy of our method in comparison with other string-matching, and graph-comparison methods used as baseline.

Extracted Key Phrases

5 Figures and Tables

Cite this paper

@inproceedings{GarcaDez2011ASW, title={A simple-cycles weighted kernel based on harmony structure for similarity retrieval}, author={Silvia Garc{\'i}a-D{\'i}ez and Marco Saerens and Mathieu Senelle and François Fouss}, booktitle={ISMIR}, year={2011} }