Date of Award

1986

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Abstract

The applications of speech synthesis for computer voice response and speech analysis present the need for highly intelligible and natural synthesized speech. In order to improve the synthesis of fricative and related sounds, the use of simple models for the source spectrum of fricative sounds is investigated. The investigation is based on the use of a vocal tract analog and experimental measurements. Measurements of the sound pressure spectra of fricative consonants are made. Simple sound pressure measurements and measurements based on the technique for measuring intensity are utilized. The fricatives studied are /f/, /th/, /s/, /sh/, and /h/. Fricative sound source spectra are determined by applying an inverse filter to the measured fricative sound pressure spectra. The inverse filtering function is derived from a vocal tract analog. The resulting fricative source spectra are fit to a truncated Fourier series. The results show that structure is evident in all the source spectra except /f/. The presence of structure was related to turbulent flows. The structure of turbulent flows is relevant since fricative sound production is induced by turbulence. The structure of turbulent flows with Reynolds number near the critical Reynolds number is dependent on the initial conditions, the boundary conditions, and on the nonlinearity of the Navier Stokes equations. These three factors are tied together by bifurcation theory which is used to explain the structure present in the fricative source spectra. Also, the possibility that the structure is a by-product of the vocal tract analog is allowed. In any case, the structure evident in the source spectra indicates the use of simple models for the source spectra of fricative sounds is in error or the vocal tract analog requires revision. The fricative source spectra determined in this study can be used in future speech synthesizers. Also, the same procedure employed in this study can be used for speech analysis of speech impaired subjects.

Pages

190

Share

COinS