Date of Award

Summer 1982

Document Type

Thesis - Restricted

Degree Name

Master of Science (MS)

Department

Electrical and Computer Engineering

Abstract

Two real-time continuous speech categorization systems were implemented, optimized, and tested. The first of these systems is a new speech categorization system which makes independent voice/no-voice and friction/no-friction classifications. Each decision is based on a measure of the absolute energy in a given frequency range of normalized speech. The second system, which is based on the spectral energy distribution of continuous speech, makes a voice/unvoice classification. The performance of these two systems was compared in a very straightforward and unbiased manner. The speech data base that was used to test each system, was manually classified using an interactive speech editor. An improved definition of performance accuracy was used to compute, and subsequently compare, the performance of the two speech categorization systems. The performance of each speech categorization system along with some important statistics of the speech data base are presented.

Share

COinS

Restricted Access Item

Having trouble?