Fast match acoustic models in large vocabulary continuous speech recognition

Author

Garudadri, H. ; Labute, P. ; Boulianne, C. ; Kenny, P.

Author_Institution

INRS Telecommun., Vardun, Que., Canada

Volume

ii

fYear

1994

fDate

19-22 Apr 1994

Abstract

First pass acoustic models in a two pass, large vocabulary, speaker dependent, continuous speech recognizer are discussed. The models are based on single state VQ, compared to our earlier 5-state continuous HMMs. The loss in performance was recovered using context dependent models, multiple codebooks and improvements in the computation of the observation vector. This resulted in a first pass inclusion of >97% while reducing the computational burden by a factor of 30. Currently, the system runs real-time on an AT&T DSP32C module and an HP-735 workstation, with recognition accuracies of up to 90% for a 5000 word vocabulary task

Keywords

digital signal processing chips; speech coding; speech recognition; vector quantisation; vocabulary; workstations; AT&T DSP32C module; HP-735 workstation; context dependent models; continuous speech recognizer; fast match acoustic models; first pass acoustic models; large vocabulary; large vocabulary continuous speech recognition; multiple codebooks; observation vector; real-time system; recognition accuracies; single state VQ; speaker dependent recognition; Business; Cepstral analysis; Context modeling; Hidden Markov models; History; Loudspeakers; Mel frequency cepstral coefficient; Real time systems; Speech recognition; Vocabulary;

fLanguage

English

Publisher

ieee

Conference_Titel

Acoustics, Speech, and Signal Processing, 1994. ICASSP-94., 1994 IEEE International Conference on

Conference_Location

Adelaide, SA

ISSN

1520-6149

Print_ISBN

0-7803-1775-0

Type

conf

DOI

10.1109/ICASSP.1994.389695

Filename

389695