Date of Award


Document Type

Thesis - Restricted

Degree Name

Master of Science (MS)


Electrical and Computer Engineering

First Advisor

Niederjohn, Russell


The processing of speech signals corrupted by noise is an active area of research at Marquette University. The work presented here supports the on going investigation into formant frequency estimation from noise corrupted signals. The work included development of an interactive program to support the generation of a formant standards data base and the use of this data base to test the performance of two formant estimation algorithms when the speech signal under analysis has been corrupted by additive white noise. The program for formant standard generation, FORSTAND, provides semi-automatic formant extraction for a speech utterance of the Marquette speech data base. The program initially performs automatic estimation of the formant locations from an estimated spectrum of the vocal tract. The user can review and modify the automatic selections. The formant estimation algorithms tested for their performance on noise corrupted speech were the zero-crossing consistancy algorithm developed by M. Lahat [19 ], and a fixed segment algorithm based on linear predicative coding [3]. The linear predicative coding algorithm displayed better performance when the noise levels were low. However, when the noise levels were very high the zero-crossing consistancy algorithm was more accurate.



Restricted Access Item

Having trouble?