Automatic recognition of affect and emotion in speech

Research on emotional speech is no longer a field dealt with just in psychology or linguistics. This topic has gained increasing interest in the engineering and computer science community because it has been shown that computers can recognise some affective states in the speech signal. In this thesis the concept to recognise affect and emotion in speech is examined. Acoustical information on the utterance level is used but no lexical cues. First a theory of emotions and some acoustical correlates of emotional speech are reviewed. Then a computational emotion recogniser is designed and implemented in the Java programming language. When it is tested on emotional speech recorded by actors, the accuracy to discriminate the states of calm and agitation is about 97 %.