ASA 129th Meeting - Washington, DC - 1995 May 30 .. Jun 06

1aSC30. Removal of phase distortion from vowel-like signals.

Kenton A. Shipley

Casey R. Winkel

P. G. Vaidya

Dept. of Mech. and Mater. Eng., Washington State Univ., Pullman, WA 99164

In this paper, specific types of signals are considered. To generate these signals, we begin with x(t)=A cos[wt+(phi)(t)], where A and w are constants, but (phi)(t) is a function of time generated by a random walk. The other step to generate the final signal is to pass x(t) through a nonlinear filter such that, were (phi)(t) to be constant the output would be of simple harmonics at w, 2w, 3w, etc. However, since x(t) is nonstationary, the output signal, say y(t), would also be nonstationary. It would be shown that human vowel signals seem to follow this nonlinear pattern. It is of interest to remove (or modify) the phase distortion due to the randomness of (phi)(t). This has been accomplished by using two different methods. One uses the trans-spectral coherence technique [P. G. Vaidya and M. J. Anderson, 2370--2378 (1991)] and the other uses the complex FFT components in a manner similar to a rigid body rotation in dynamics. These methods have been used to produce both distortion-free and highly modulated versions from a given human vowel signal. [sup a)]Now at Martin Marietta, Knolls Atomic Power Labs, Schenectady, NY.