Music Similarity Measures: What’s the Use?

By Jean-Julien Aucouturier and François Pachet.

Proceedings of the 3rd International Conference on Music Information Retrieval (ISMIR 2002), pp. 157–163, Paris, France. DOI: 10.5281/zenodo.1418257.

2026 ISMIR Test-of-Time Honorable Mention

Read the paper (PDF) · Proceedings record (DOI)

A conversation with Jean-Julien Aucouturier

Watch on YouTube

About the paper

This paper asks how timbre can help listeners find music that sounds similar. It introduces a way to compare the overall sound of recordings using statistical models of audio features and shows how timbre can complement other notions of similarity in music browsing.

Abstract

Electronic Music Distribution (EMD) is in demand of robust, automatically extracted music descriptors. We introduce a timbral similarity measures for comparing music titles. This measure is based on a Gaussian model of cepstrum coefficients. We describe the timbre extractor and the corresponding timbral similarity relation. We describe experiments in assessing the quality of the similarity relation, and show that the measure is able to yield interesting similarity relations, in particular when used in conjunction with other similarity relations. We illustrate the use of the descriptor in several EMD applications developed in the context of the Cuidado European project.

Related work

Hubness in Music Similarity follows this work from the initial timbre measure to the later studies of recurring false matches.

The follow-up Improving Timbre Similarity: How High’s the Sky? systematically tests variations of this measure and examines where the approach reaches its limits.

A Scale-Free Distribution of False Positives for a Large Class of Audio Similarity Measures (2008) investigates why some songs repeatedly appear as false matches in timbre-based music searches.

Back to Publications