Vocal Tract Modelling and Speech Synthesis: Its Use for Improving Knowledge on Speech Production

Abstract

Publisher Summary During the past 20 years, significant progress in speech synthesis can be associated with better vocal-tract modeling. A model is constructed from all available data coming from acoustical, physiological, and neurophysiological studies. Such a model can be improved by using analysis by synthesis techniques. The differences between behaviors of the original system and the model being detected, new problems can be formalized and, perhaps, solved. Many back and forth motions will be needed to postulate a reasonable model of the speech production, that is, a model quite usable for speech synthesis. Obviously, with a correct model, better knowledge about speech production can be obtained. For example, the function and the importance of any element of the model can be studied much more easily than in the human system. At first, a direct simulation of the vocal tract as an acoustic tube (transmission-line analogs of the vocal tract) was studied. After that, formant synthesizers (terminal analog synthesizers) which are much easier to control, were used for numerous studies on speech synthesis, synthesis by rules and for perception tests. At the present time, research using analogs of the vocal tract are again being undertaken because this kind of model is more adapted for basic research: on losses along the vocal tract, on the coupling effects between vocal tract and the source, that between vocal tract and nasal tract, and such. In other respects, it is also necessary to have a right model of the vocal cords, the voice source seeming to be responsible in part for the quality of the synthesized speech.