Classification of emotions in vocal responses

Çağlayan, Ece
Emotion is a relatively short-term conscious experience characterized by intense mental activity and high level of pleasure or dissatisfaction. During a dialogue, a person feels the emotion in the other person voice and chooses accordingly how to react. Within the scope of this thesis, it is investigated whether we can distinguish the emotional content of a response from the speech signals regardless of the semantics. Accordingly, audio recordings containing six basic and neutral emotions were played to the participants severally. Since the aim is to measure the effect of the acoustic structure rather than semantic structure we took account of German voice recordings from the Berlin emotional speech database. In this respect, meaningful Turkish sentences comprising neutral words were shown on the screen randomly as the next step of the experiment. Participants were expected to read these sentences with their emotional reaction to the previous voice record. Audio recordings of the participants were taken. Thus, an artificial dialogue was reproduced. To our knowledge, this is the first research of classification of emotional responses to an emotional audio record. In our study, 30 basic features were extracted from speech records of 21 subjects who participated in our experiment and their emotional responses to audio records were classified using an artificial neural network. By this way, it is considered that the measurement of the acoustic response to a particular emotion can be classified. After the statistical analysis, it has been shown that the response given for the anger can be classified in reasonable rate. In addition to classifying the responses to emotional audio records, we foresee that classification performance for emotional responses can be increased. 
Citation Formats
E. Çağlayan, “Classification of emotions in vocal responses,” M.S. - Master of Science, Middle East Technical University, 2017.