Abstract

Music has been shown to impact the affective states of the listener. The emotion in music task at the MediaEval challenge 2015 focuses on predicting the affective dimensions of valence and arousal in music using low level features. In particular, this edition of the challenge involves prediction on full length songs given a training set containing smaller 30 second clips. We approach the problem as a regression task and test several regression algorithms. We proposed these regression methods on the dataset from previous edition of the same task (Mediaeval 2014) involving prediction on 30 second clips instead of full length songs. Through evaluation on the 2015 data set, we obtain a point of reference for the model performances on longer song clips. Whereas our models perform relatively well in predicting arousal (root mean square error:. 24), we do not obtain good results for valence prediction (root mean square error:. 35). We analyze the results and the experimental setup and discuss plausible solutions for a better prediction.

Date

2015

Authors

Rahul Gupta, Shrikanth S Narayanan

Conference

MediaEval