Late integration of features for acoustic emotion recognition

It is widely accepted that the ability to understand emotion or affect from speech is central to the design of more natural human-computer interfaces. This paper explores the classification of natural emotional speech along four affective dimensions, using hidden Markov models (HMMs). A number of features are tested, some of which have never before been applied to emotion recognition. Finally, these different features are combined discriminatively to achieve a competitive performance on the AVEC 2011 affect classification task [1].