The TUM Approach to the MediaEval Music Emotion Task Using Generic Affective Audio Features
Felix Johannes Weninger, Florian Eyben, Björn Wolfgang Schuller · 2013
This paper describes the TUM approach for the MediaEval Emotion in Music task which consists of non-prototypical music retrieved from the web, annotated by crowdsourcing. We use Support Vector Machines and BLSTM recurrent neural networks for static and dynamic arousal and valence regression. A generic set of acoustic features is used that has been proven effective for affect prediction across multiple domains. In the result, the best models explain 64 and 48 % of the annotations ’ variance for arousal and valence in the static case, and an average Kendall’s tau with the songs’ emotion contour of.18 and.12 is achieved in the dynamic case. 1.