Feature sets for automatic classification of dimensional affect

Automatic recognition of emotion from speech has many potential applications, from the design of more user friendly human-machine interfaces to the improvement of speech recognition for natural speech. As this is a relatively young field there remains uncertainty in the literature over the best classifier architectures and feature sets for emotion classification. In this work we explore the classification of emotion from speech using Hidden Markov Models (HMMs). We show that using HMM classification we can significantly reduce the size of our feature set from that proposed in the recent Audio/Visual Emotion Challenge (AVEC 2011), while maintaining a performance similar to that of the winning classifier from the audio sub-challenge of the AVEC challenge. We compare the performance of our HMM classifier using five different feature sets, and show that for dimensional classification the optimum feature set is dependent on the emotion dimension in question.