Emotional Speech Synthesis Based On The Curve Regression

Speech technology has played a very important role in today’s rapid development of interaction. With the growing requirement of the computer, the research and the application of emotional speech have become increasingly mature. Setting up of a high quality of emotional speech database provides an important basis for studying of emotional speech. For the problem of establishment of emotional speech database and emotional speech synthesis, this paper carried out researches into the following points:Through the studying and research of emotional theory system, combine the actual needs of our laboratory’s studying, this paper determines that there are four types of emotions in TYUT emotional speech database:neutral, joy, anger, surprise. Referring to the existing theory of emotional speech database and building process, TYUT emotional speech database’s specific process is designed:text determining and selection with the text filtering method; through the way that combines induction of images and music to record voice; screening the statements through the method that combines subjective hearing with recognition system. TYUT emotional speech database’s establishment provides an important experimental basis for the next research of emotional speech.This paper introduces the SPSS statistical analysis software for the first time to the study of speech parameters. According to the analysis of parameter characteristics that influences the emotional of the detailed statement, the curve regression analysis of parameters among the different emotional speeches is set up. The establishment of a reasonable curve regression model helps a lot for studying the parameters characteristics, and provides a good foundation to realize the emotional speech synthesis.Using the curve regression model of parameter characteristics, metrical parameters of neutral statements in the database are changed. And then emotional statements are synthesized. By comparing emotional statements in database and synthesized emotional statement, combined with the subjective hear experiment, verify the validity of the curve regression model.