Extracting action-sound features from a sound-tracing study

Abstract

The paper addresses possibilities of extracting information from music-related actions, in the particular case of what we call sound-tracings. These tracings are recordings from a graphics tablet of subjects' drawings associated with a set of short sounds. Although the subjects' associations to sounds are very subjective, and thus the resulting tracings are very different, an attempt is made at extracting some global features which can be used for comparison between tracings. These features are then analyzed and classified with an SVM classifier.