Hybrid simulation of speech waveforms utilizing a Gaussian wave function representation

Abstract

Most of the prior research work concerned with the analysis and synthesis of speech has employed Fourier series and transforms as the mathematical framework for the characterization of live speech. Current studies at UCSB have taken a new approach, namely the time domain analysis of speech waveforms. This work, per formed under the direction of Dr. Glen J. Culler, is based on a Gaussian wave-function approach to the analysis, synthesis and decomposition of speech elements. This paper is concerned with the hybrid simulation of speech waveforms utilizing the wave-function representation. A speech simulator consisting of a hybrid interconnec tion of four TR-20 analog computers is described. Re sults of the simulation of two vowel sounds, the "o" in "on" and the "a" in "at," are presented and compared with the original speech data. The results form an im pressive check on the wave-function approach and have provided the impetus for the realization of a hybrid speech-synthesizer based on this technique.