Improved speaker adaptation using speaker dependent feature projections

Author

Matsoukas, Spyros ; Schwartz, Richard

Author_Institution

BBN Syst. & Technol. Corp., Cambridge, MA, USA

fYear

2003

fDate

30 Nov.-3 Dec. 2003

Firstpage

273

Lastpage

278

Abstract

We extend the formulation of constrained maximum likelihood linear regression (CMLLR) adaptation to take into account full covariance matrices in the adapted model, and we use it in conjunction with heteroscedastic linear discriminant analysis (HLDA) in order to estimate speaker dependent feature projections on both training and test data. Results on the broadcast news corpus show that the proposed HLDA adaptation technique is very effective, even when combined with traditional CMLLR and MLLR adaptation, providing up to 8% relative improvement in recognition accuracy.

Keywords

covariance matrices; maximum likelihood estimation; speech recognition; broadcast news corpus; constrained MLLR adaptation; constrained maximum likelihood linear regression adaptation; covariance matrices; heteroscedastic LDA adaptation; heteroscedastic linear discriminant analysis adaptation; speaker adaptation; speaker dependent feature projection estimation; Acoustic testing; Broadcasting; Covariance matrix; Hidden Markov models; Linear discriminant analysis; Linear regression; Loudspeakers; Maximum likelihood linear regression; Speech; State estimation;

fLanguage

English

Publisher

ieee

Conference_Titel

Automatic Speech Recognition and Understanding, 2003. ASRU '03. 2003 IEEE Workshop on

Print_ISBN

0-7803-7980-2

Type

conf

DOI

10.1109/ASRU.2003.1318453

Filename

1318453