A multiple feature model for musical similarity retrieval

Abstract

Automated musical similarity search and retrieval has gained great attention in recent years, as testified by the number of proposed approaches on this topic. Despite the ``fuzzy'' nature of similarity, which varies from one person to another, perceptual low level features combined with appropriate classification schemes have proven to perform satisfactorily for this task. While a single feature only captures some selective characteristics of an audio signal, this information may, in some cases, not be sufficient to properly identify similarities between songs. This paper presents a system which combines a set of acoustic features for the task of retrieving similar sounding songs in a database with a low computational cost. The methodology for optimum feature selection and combination is explained, as well as the classification process. Finally, the results of a subjective listening test, aimed at assessing the system's performance, are presented and discussed. Some interesting applications of such a system are described.