MusicBox : navigating the space of your music

Abstract

Navigating increasingly large personal music libraries is commonplace. Yet most music browsers do not enable their users to explore their collections in a guided and manipulable fashion, often requiring them to have a specific target in mind. MusicBox is a new music browser that provides this interactive control by mapping a music collection into a two-dimensional space, applying principal components analysis (PCA) to a combination of contextual and content-based features of each of the musical tracks. The resulting map shows similar songs close together and dissimilar songs farther apart. MusicBox is fully interactive and highly flexible: users can add and remove features from the included feature list, with PCA recomputed on the fly to remap the data. MusicBox is also extensible; we invite other music researchers to contribute features to its PCA engine. A small user study has shown that MusicBox helps users to find music in their libraries, to discover new music, and to challenge their assumptions about relationships between types of music.