Date of Award

Spring 1992

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Electrical & Computer Engineering

Program/Concentration

Electrical Engineering

Committee Director

David Livingston

Committee Member

Stephen Zahorian

Committee Member

Jack Stoughton

Call Number for Print

Special Collections LD4331.E55K41

Abstract

Formants are the natural frequencies of the human vocal tract. Existing methods for estimating formants from speech signals are computationally complex and subject to errors for certain type of speech sounds. This thesis describes a method for estimating vowel formant frequencies from Discrete Cosine Transform Coefficients (DCTC's), a form of cepstral coefficients, using a feedforward neural network with back-propagation training. Experimental results are based on a large multispeaker data base. The results are obtained for both a linear transformation and a feedforward neural network with a nonlinear hidden layer. In general, the neural network transformation is superior to the linear transformation for formant estimation. Thus our experiments indicate the nonlinear nature of the relationship between DCTC's and formants. However, since the results are always much better for training data as compared to test data, a large data base is necessary for adequate neural network training. Vowel classification experiments show that estimated formants can discriminate vowels nearly as well as 14 DCTC's.

Rights

In Copyright. URI: http://rightsstatements.org/vocab/InC/1.0/ This Item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s).

DOI

10.25777/em4k-jh31

Download

Share

COinS