Towards intuitive speech interaction by the integration of emotional aspects

In this paper new results in instantaneous recognition of emotion in non-verbal speech are presented. As classification method dynamic programming with dynamic time warp or Bakis-hidden-Markov-models with vector quantization or Gaussian mixtures are used to analyze the pitch and energy contour of a speech signal. As emotional states joy, anger, fear, sadness, disgust, irritation, and an additional neutral user state have been evaluated. As rather unusual innovative user states the influences of tiredness and alcohol consumption of a speaker on his speech have been analyzed by use of the same methods. One of the main goals of the presented work is to keep the models applicable for new users and to keep recognition simple for real-time evaluation. Finally the observed results are presented and discussed.