Analyzing And Modeling Voice Quality And Jitter In Emotional Speech Synthesis

With the development of speech science and computer technology, the speech synthesis system nowadays could synthesize speech with high intelligibility. However, there are still large gap between these machine-generated utterances and human natural speech, which is more expressive and spontaneous than the synthesized narrative speech. So the expressive speech synthesis urges researchers to reconsider these systems and to propose some new models for describing the subtle change in emotional and expressive speech.There are a great many factors to affect the natureness of synthesized speech, for example, voice quality class, metrical class and articulation class. What is the crucial part to improve the current system is to find ways rebuild these models. In this thesis, the author mainly analyzes these factors from voice quality and jitter variance aspects. There are three major components in this paper:1, the paper shows the basic concept of expressive speech synthesis and then gives a brief analysis on physiological and perceptive facet. Different voice quality parameters modeled by LF model are gained with the inverse filtering technique.2, the paper proposes a new method of separating the two components in the total jitter: random jitter and deterministic jitter.3, the paper also analyzes the jitter's pattern in different emotions, different voice qualities, and different tune in Standard Chinese.We wish these results would be useful in the further researchs of expressive speech synthesis.