Discovering Music Structure via Similarity Fusion

Automatic methods for music navigation and music recommendation exploit the structure in the music to carry out a meaningful exploration of the “song space”. To get a satisfactory performance from such systems, one should incorporate as much information about songs similarity as possible; however, how to do so is not obvious. In this paper, we build on the ideas of the Probabilistic Latent Semantic Analysis (PLSA) that have been successfully used in the document retrieval community. Under this probabilistic framework, any song will be projected into a relatively low dimensional space of “latent semantics”, in such a way that all observed similarities can be satisfactorily explained using the latent semantics. Therefore, one can think of these semantics as the real structure in music, in the sense that they can explain the observed similarities among songs. The suitability of the PLSA model for representing music structure is studied in a simplified scenario consisting of 4412 songs and two similarity measures among them. The results suggest that the PLSA model is a useful framework to combine different sources of information, and provides a reasonable space for song representation.