A video compression-based approach to measure music structural similarity
Résumé
The choice of the distance measure between time-series representations can be decisive to achieve good classification results in many content-based information retrieval applications. In the field of Music Information Retrieval, two-dimensional representations of the music signal are ubiquitous. Such representations are useful to display patterns of evidence that are not clearly revealed directly in the time domain. Among these representations, self-similarity matrices have become common representations for visualizing the time structure of an audio signal. In the context of organizing recordings, recent work has shown that, given a collection of recordings, it is possible to to group performances of the same musical work based on the pairwise similarity between structural representations of the audio signal. In this work, we introduce the use of the Campana- Keogh distance, a video compression-based measure, to compare musical items based on their structure. Through extensive experiments, we show that the use of this distance measure outperforms the results of previous work using similar approaches but other distance measures. Along with quantitative results, detailed examples are provided to to illustrate the benefits of using the newly proposed distance measure.
Origine : Fichiers produits par l'(les) auteur(s)
Loading...