Analysis and synthesis of emotional voice by time-frequency method

This paper proposes an analysis method for emotional voice and application to emotional voice synthesis in man-machine interaction. The analysis is based on a high resolution time-frequency method and the clustering technique of four types of basic emotion. Based on the emotional analysis, a voice with the neutral emotion is transformed to the particular emotional voice according to time-frequency emotional modifications. In the simulations, four types of emotional voices (neutral, anger, sadness, joy) are analyzed using 160 samples of speech signals. About 88 (%) of average discrimination rate is achieved. Based on the cluster analysis results, an unknown voice is also discriminated to show the effectiveness of our method. Further, we attempt to synthesize an emotional voice.