Emotion modeling in speech production using emotion space

This paper describes the modeling scheme of emotions appearing in a speech production by using a neural network and the synthesizing technique of emotional condition from neutral speech. To model emotion conditions in speech production, emotion space is introduced. Emotion space can represent the emotion condition appearing in speech production in a two dimensional space and realize both mapping and inverse mapping between the emotion condition and the speech production. We developed the emotional speech synthesizer to synthesize emotional speech. The emotional speech synthesizer has an ability to synthesize emotional speech by modifying neutral speech in its timing, pitch and intensity. This paper also describes the subjective evaluation result of synthesized speech from the emotion space.