Music Icons: Procedural Glyphs for Audio Files

Abstract

Nowadays, a personal music collection may comprise thousands of MP3 files. Visualization can help the user to gain an overview and to find similar songs inside so large a set. We describe a method to create icons from audio files in such a way that songs which the user considers similar receive similar icons. This allows visual data mining in standard directory listings of window-based operating systems. The icons consist of bloom-like shapes, whose form and color depend on eight parameters. These parameters are controlled through a neural net, the input of which are audio features that are extracted algorithmically from the MP3 files. To adapt the system to the user's perception and interests, the neural net is initially trained with a small set of songs and icons. User studies done on the system demonstrate a strong perceptual relation between music and icons