Prediction of Time-varying Musical Mood Distributions from Audio

The appeal of music lies in its ability to express emotions, and it is natural for us to organize music in terms of emotional associations. But the ambiguities of emotions make the determination of a single, unequivocal response label for the mood of a piece of music unrealistic. We address this lack of specificity by modeling human response labels to music in the arousal-valence (A-V) representation of affect as a stochastic distribution. Based upon our collected data, we present and evaluate methods using multiple sets of acoustic features to estimate these mood distributions parametrically using multivariate regression. Furthermore, since the emotional content of music often varies within a song, we explore the estimation of these A-V distributions in a time-varying context, demonstrating the ability of our system to track changes on a short-time basis.