UBM Based Speaker Selection and Model Re-Estimation for Speaker Adaptation

Author

Wang, Jian ; Guo, Jun ; Liu, Gang ; Lei, Jianjun

Author_Institution

Sch. of Inf. Eng., Beijing Univ. of Posts & Telecommun.

Volume

2

fYear

2006

fDate

17-19 July 2006

Firstpage

856

Lastpage

860

Abstract

Based on speaker selection, speaker adaptation technology can get a promising performance. In such system, how to represent a speaker and the computation of selection are still big issues. In this paper, we take Gaussian mixture model (GMM) as representation of a speaker, which adapted from universal background model (UBM). Likelihood ratio (LR) and cross likelihood ratio (CLR) are utilized for speaker selection. Furthermore, a single-pass re-estimation procedure, conditioned on the speaker-independent model is shown. This adaptation strategy was evaluated in a large vocabulary speech recognition task. A relative gain of 11% with respect to the baseline system is achieved

Keywords

Gaussian processes; speaker recognition; Gaussian mixture model; cross likelihood ratio; speaker adaptation; speaker representation; speaker selection; speech recognition; universal background model; Cognitive informatics; Hidden Markov models; Loudspeakers; Maximum likelihood linear regression; Speech recognition; Statistical analysis; Statistical distributions; Telecommunication computing; Testing; Vocabulary; UBM; speaker adaptation; speaker selection;

fLanguage

English

Publisher

ieee

Conference_Titel

Cognitive Informatics, 2006. ICCI 2006. 5th IEEE International Conference on

Conference_Location

Beijing

Print_ISBN

1-4244-0475-4

Type

conf

DOI

10.1109/COGINF.2006.365603

Filename

4216521