Physiologically based speech synthesis using digital waveguide model

Abstract

Problem in the area of speech synthesis research is that synthetic speech generally sounds unnatural. Artificially generated speech lacks the smoothness of natural speech, and this in turn can hinder intelligibility. Natural sounding synthetic speech can be generated by modeling inflection, modeling the glottal source, improving techniques of concatenating segments of speech, and modeling the temporal nuances of phrasing. Computer models of the human voice-production system are an important research tool for speech scientists and engineers. This paper uses physiological data to model human vocal tract to tackle above mentioned problem. Main focus is on piece-wise cylindrical model of the human vocal tract, since it is a relatively simple (in theory), and is capable of producing good results and most natural sounding speech synthesis. And it attempts to describe the comparison between different numbers of segments used to model vocal tract. © Springer-Verlag Berlin Heidelberg 2011.