James Delmege


The objective of this thesis was to examine computer techniques for classifying speech signals into four coarse phonetic classes: vowel-like, strong fricative, weak fricative and silence. The study compared classification results from the K-means clustering algorithm using Euclidian distance measurements with classification using a multivariate maximum likelihood distance measure. In addition to the comparison of statistical methods, this study compared classification using several tree-structured decision making processes. The system was trained on ten speakers using 98 utterances with both known and unknown speakers. Results showed very little difference between the Euclidian distance and maximum likelihood; however, the introduction of the tree structure on both systems had a positive influence on their performance.

Library of Congress Subject Headings

Automatic speech recognition; Pattern recognition systems; Phonetics--Analysis--Data processing

Publication Date


Document Type


Department, Program, or Center

Computer Science (GCCIS)


Biles, John

Advisor/Committee Member

Anderson, Peter


Note: imported from RIT’s Digital Media Library running on DSpace to RIT Scholar Works. Physical copy available through RIT's The Wallace Library at: TK7882.S65 D458 1988


RIT – Main Campus