Speaker adaptation using discriminative linear regression on time-varying mean parameters in trended HMM

Author

Chengalvarayan, Rathinavelu

Author_Institution

Lucent Technol., Bell Labs., Naperville, IL, USA

Volume

5

Issue

3

fYear

1998

fDate

3/1/1998 12:00:00 AM

Firstpage

63

Lastpage

65

Abstract

In this letter, we report our recent work on applications of the combined maximum likelihood linear regression (MLLR) and the minimum classification error training (MCE) approach to estimating the time-varying polynomial Gaussian mean functions in the trended hidden Markov model (HMM). We call this integrated approach the minimum classification error linear regression (MCELR), which has been developed and implemented in speaker adaptation experiments using TI46 corpora. Results show that the adaptation of linear regression on time-varying mean parameters is always better when fewer than three adaptation tokens are used.

Keywords

hidden Markov models; maximum likelihood estimation; polynomials; speech recognition; time-varying systems; HMM; TI46 corpora; adaptation token; discriminative linear regression; maximum likelihood linear regression; minimum classification error linear regression; minimum classification error training; speaker adaptation; speech recognition; time-varying mean parameters; time-varying polynomial Gaussian mean functions; trended hidden Markov model; Acoustic testing; Adaptation model; Covariance matrix; Hidden Markov models; Linear regression; Loudspeakers; Maximum likelihood estimation; Maximum likelihood linear regression; Polynomials; Vectors;

fLanguage

English

Journal_Title

Signal Processing Letters, IEEE

Publisher

ieee

ISSN

1070-9908

Type

jour

DOI

10.1109/97.661562

Filename

661562