Improved vocabulary-independent sub-word HMM modelling

Author

Wood, Lynn C. ; Pearce, David J B ; Novello, Frederic

Author_Institution

GEC-Marconi Ltd., Wembley, UK

fYear

1991

fDate

14-17 Apr 1991

Firstpage

181

Abstract

The authors describe two techniques for improving the performance of subword recognition on open vocabularies using vocabulary-independent training. The first uses a subtriphone unit called a phonicle to allow triphones which have not been encountered in the training data to be built from contexts which have been sufficiently trained. The second uses linear discriminant analysis to improve discrimination between sound classes. The two techniques have been evaluated for speaker-dependent operation on an open vocabulary task. The recognizer is based on hidden Markov modeling (HMM) using continuous probabilities. The results obtained show that both techniques lead to improved recognition performance

Keywords

Markov processes; acoustic signal processing; speech analysis and processing; speech recognition; HMM; continuous probabilities; hidden Markov modeling; linear discriminant analysis; open vocabularies; phonicle; recognition performance; sound classes; speech analysis; subtriphone unit; subword recognition; training data; triphones; vocabulary-independent training; Clustering algorithms; Context modeling; Hidden Markov models; Linear discriminant analysis; Probability distribution; Speech analysis; Speech recognition; Testing; Training data; Vocabulary;

fLanguage

English

Publisher

ieee

Conference_Titel

Acoustics, Speech, and Signal Processing, 1991. ICASSP-91., 1991 International Conference on

Conference_Location

Toronto, Ont.

ISSN

1520-6149

Print_ISBN

0-7803-0003-3

Type

conf

DOI

10.1109/ICASSP.1991.150307

Filename

150307