A systematic assessment of the accuracy of vocal tract area function estimates made from the speech waveform

Citation

Milenkovic, Paul H. (1981) A systematic assessment of the accuracy of vocal tract area function estimates made from the speech waveform. Dissertation (Ph.D.), California Institute of Technology. doi:10.7907/6ydf-ke74. https://resolver.caltech.edu/CaltechETD:etd-09282006-135046

Abstract

By performing Linear Predictive Coding (LPC) analysis on the speech waveform, it is possible to determine the cross sectional areas, or area function, of a discrete section acoustic tube model of the vocal tract. It is a matter of controversy, however, as to whether the areas of the acoustic tube model accurately estimate the areas of the actual vocal tract. There are several sources of error which cause the estimated areas to differ from the true areas. A procedure for estimating the spectrum of the vocal tract response in terms of LPC derived formant frequencies and bandwidths is discussed; the areas of the acoustic tube model can be calculated from these frequency and bandwidth values. The accuracy with which formant frequencies and bandwidths can be estimated is evaluated by experiments where the frequency and bandwidth of a one resonator vocal tract model are estimated. The accuracy of the complete procedure for estimating the area function from speech is evaluated by experiments where the area function is estimated from synthetic speech sounds. These speech sounds are synthesized from known vocal tract shapes against which the estimated area function can be compared.

Thesis Files