Applications of HMMs for the Recognition of Emotional Sequences in the Valence-Arousal Space

Authors: David Hübner, Ronald Böck, Andreas Wendemuth

Abstract:

This paper will show how models can be generated, which are capable of recognizing sequences of emotional states from speech. For this purpose Hidden Markov Models (HMMs) are introduced, which are trained on spontaneous, nonacted emotions. Unlike other publications in this area, whose main focus is often on the classification in one of the basic classes introduced by Ekman or Plutchik, we will generate 2-dimensional representations of the user’s emotion in the Valence- Arousal space. Hence, not only the basic emotions are recognized, but also an additional parameter, the word frequency, is extracted from the speech signal. We trained two gender specific models and one combined model and tested these afterwards on unknown data. The evaluation of the robustness is done by using two cross-validation methods.