Expressive Speech Synthesis System Using Unit Selection

Abstract

Abstract. Speech for realistic environment is hard to achieve. Emotion synthesizing is one way to achieve realistic and natural sounding speech. Use of right emotion in synthesized speech generates the speech which is more effective and natural for listener. The implementation of emotions is very difficult, as word "emotion" has no single definition. There have been various attempts in creating emotional speech synthesis but perfect or near to ideal system has not been developed so far. Our paper is an attempt to create emotional speech synthesizer, where we have used the emotional database recorded in our own voice. We have used unit selection and CART method to implement it. We have taken class room environment for teaching pre-school students with three emotions i.e neutral, happy, sad and tested our synthesizer with twenty listeners and found that listeners have significantly identify the emotional state of speaker.