USE OF FUNDAMENTAL FREQUENCIES SHAPED BY GENERATION PROCESS MODEL FOR HMM-BASED SPEECH SYNTHESIS

Generation process model of fundamental frequency (F-0) contours is known to represent global movements of F-0's keeping a clear relation with linguistic information of utterances. While HMMbased speech synthesis can generate a good quality of speech, problems, which arise from frame-by-frame processing, are pointed out. These problems are expected to be solved by incorporating the model constraints. A method is developed to use F-0 contours approximated by the model for HMM training instead of observed F-0 contours. A clear improvement in the quality of synthetic speech is shown through listening experiments. In the method, fragments of F-0 contours not represented by the model (F-0 residuals) are ignored. A scheme is further introduced to cope with the issue; F-0 residuals are also included in the training and synthesis processes of HMM-based speech synthesis, and the generated F-0 residuals are added to the model-based Fo's before the waveform generation. The model constraint has another merit; relations between generated F-0 contours and texts are clear, and it is possible to add linguistic information such as emphasis to synthetic speech, or to change speaking styles through manipulating Fo's in the F-0 model framework. Several experimental results supporting the advantages of the method are shown.