Abstract

In building a speaker independent continuous speech understanding sys tem, a method is needed to translate the speech signal to a phoneme string. Current recognition attempts based on statistical methods are not as accurate as desired and require vast numbers of templates for comparison. On the other hand, human experts can transcribe phoneme strings from spectrograms with a high degree of accuracy, without needing to match unidentified seg ments against identified templates. This paper discusses the development of Fric, an expert system for identifying fricative phonemes, intended to func tion as part of a phoneme identification knowledge source. Fric attempts to mimic the techniques used by human transcribers in identifying phonemes. RuleMaster, an expert system building tool developed by Radian Corporation, was used to create this system. Fric uses both forward and backward chaining as control strategies and includes an explanation system for both debugging and explanatory purposes.

Library of Congress Subject Headings

Automatic speech recognition; Speech processing systems; Expert systems (Computer science); Natural language processing (Computer science)

Publication Date

1987

Document Type

Thesis

Department, Program, or Center

Computer Science (GCCIS)

Advisor

Biles, John

Advisor/Committee Member

Wolf, Walter

Advisor/Committee Member

Hillenbrand, James

Comments

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.S65A84 1987

Campus

RIT – Main Campus

Download

Share

COinS