Recent experiments in large vocabulary conversational speech recognition

Author

Billa, J. ; Colhurst, T. ; El-Jaroudi, A. ; Iyer, Ravishankar ; Ma, K. ; Marsoukas, S. ; Quillen, C. ; Richardson, F. ; Siu, M. ; Zavaliagkos, G. ; Gish, H.

Author_Institution

BBN Technol., Cambridge, MA, USA

Volume

1

fYear

1999

fDate

15-19 Mar 1999

Firstpage

41

Abstract

This paper describes the improvements that resulted in the 1998 Byblos large vocabulary conversational speech recognition (LVCSR) system. Salient among these improvements are: improved signal processing, improved hidden Markov model (HMM) topology, use of quinphone context, introduction of diagonal speaker adapted training (DSAT), incorporation of variance adaptation in the MLLR framework, improvements in language modeling, increase in lexicon size and combination of multiple systems. These changes resulted in about a 7% absolute reduction in word error rates on a balanced Switchboard/Callhome English test set

Keywords

hidden Markov models; natural languages; speech recognition; HMM topology; Hidden Markov Model; Switchboard/Callhome English test set; diagonal speaker adapted training; language modeling; large vocabulary conversational speech recognition; lexicon size; multiple systems; quinphone context; signal processing; variance adaptation; word error rate reduction; Context modeling; Error analysis; Hidden Markov models; Maximum likelihood linear regression; Natural languages; Signal processing; Speech recognition; Testing; Topology; Vocabulary;

fLanguage

English

Publisher

ieee

Conference_Titel

Acoustics, Speech, and Signal Processing, 1999. Proceedings., 1999 IEEE International Conference on

Conference_Location

Phoenix, AZ

ISSN

1520-6149

Print_ISBN

0-7803-5041-3

Type

conf

DOI

10.1109/ICASSP.1999.758057

Filename

758057