On speaker-independent, speaker-dependent, and speaker-adaptive speech recognition

Author

Huang, Xuedong ; Lee, Kai-Fu

Author_Institution

Sch. of Comput. Sci., Carnegie Mellon Univ., Pittsburgh, PA, USA

Volume

1

Issue

2

fYear

1993

fDate

4/1/1993 12:00:00 AM

Firstpage

150

Lastpage

157

Abstract

The DARPA Resource Management task is used as a domain for investigating the performance of speaker-independent, speaker-dependent, and speaker-adaptive speech recognition. The error rate of the speaker-independent recognition system, SPHINX, was reduced substantially by incorporating between-word triphone models additional dynamic features, and sex-dependent, semicontinuous hidden Markov models. The error rate for speaker-independent recognition was 4.3%. On speaker-dependent data, the error rate was further reduced to 2.6-1.4% with 600-2400 training sentences for each speaker. Using speaker-independent models, the authors studied speaker-adaptive recognition. Both codebooks and output distributions were considered for adaptation. It was found that speaker-adaptive systems outperform both speaker-independent and speaker-dependent systems, suggesting that the most effective system is one that begins with speaker-independent training and continues to adapt to users

Keywords

hidden Markov models; speech recognition; DARPA Resource Management task; HMM; SPHINX; between-word triphone models; codebooks; dynamic features; error rate; hidden Markov models; output distributions; speaker dependent recognition; speaker-adaptive recognition; speaker-independent recognition; speech recognition; Computer science; Databases; Error analysis; Hidden Markov models; Resource management; Speech recognition; Testing; Training data; US Department of Defense; Vocabulary;

fLanguage

English

Journal_Title

Speech and Audio Processing, IEEE Transactions on

Publisher

ieee

ISSN

1063-6676

Type

jour

DOI

10.1109/89.222875

Filename

222875