New developments in the Lincoln stack-decoder based large-vocabulary CSR system

Author

Paul, Douglas B.

Author_Institution

Lincoln Lab., MIT, Lexington, MA, USA

Volume

1

fYear

1995

fDate

9-12 May 1995

Firstpage

45

Abstract

The system described here is a large-vocabulary continuous-speech recognition (CSR) system developed using the ARPA Wall Street Journal (WSJ) and North American Business (NAB) databases. The recognizer uses a stack decoder-based search strategy, with a left-to-right stochastic language model. This decoder has been shown to function effectively on 56 K-word recognition of continuous speech. It operates left-to-right and can produce final textual output while continuing to accept additional input. The recognizer also features recognition-time adaptation to the user´s voice. The new system showed a 48% reduction in the word error rate over the previously reported Nov. 92 system

Keywords

Bayes methods; decoding; hidden Markov models; speech recognition; ARPA Wall Street Journal database; Bayesian smoothing; HMM; Lincoln stack-decoder based large-vocabulary CSR system; North American Business database; continuous-speech recognition system; left-to-right stochastic language model; recognition-time adaptation; search strategy; word error rate; Decoding; Error analysis; Gaussian processes; Hidden Markov models; Laboratories; Notice of Violation; Paramagnetic resonance; Quantization; Reactive power; Spatial databases; Speech recognition; Stochastic processes; Wiener filter;

fLanguage

English

Publisher

ieee

Conference_Titel

Acoustics, Speech, and Signal Processing, 1995. ICASSP-95., 1995 International Conference on

Conference_Location

Detroit, MI

ISSN

1520-6149

Print_ISBN

0-7803-2431-5

Type

conf

DOI

10.1109/ICASSP.1995.479269

Filename

479269