music visualisation, semantic descriptors, music information research, information visualisation.

Can a picture accurately represent the contents of a song? In this work we introduce SongVis: an interactive, web-based, visualisation tool that represents semantic features of music. Music, despite being an aural phenomenon, is often depicted by visual means, from the ancient Greeks associating sounds with colors to virtual reality tools nowadays. To develop SongVis we considered: which features are the most important, how to extract them from music files, how to graphically represent these features, how to assemble them into a unique visualisation, how interactions could yield insights and, finally, how successful was this work considering the views of the general public. SongVis allows the comparison of songs by the inspection of their visuals, and this quality can be exploited at tasks such as: browsing music libraries for similar tracks, filtering queries based on visual features, management of music collections or playlists according to the desired visual attributes, etc. Hence, this work looks forward to benefiting the Music Information Research (MIR) field, the Information Visualisation (InfoVis) field and users in general, by providing a useful tool and a step forward at the representation of music semantics.