Investigating emotional speech parameters for speech synthesis

In this paper some of the characteristics of human emotional speech are investigated and the analysis results are reported. The collection of speech material to be analyzed and the analysis procedure were done with respect to the perspective of applying the findings to the formant based Text-to-Speech system developed in our laboratory. More specifically the following parameters have been investigated: pitch contour, intensity, speech rate and phoneme duration for four different emotions: anger, fear, joy and grief. The results of the analysis of the emotional speech data are compared to the corresponding parameters of emotionally neutral human speech. The diversion from the emotionally neutral parameters´ values indicates the modification factor necessary for the production of emotional synthetic speech