Product of Gaussians and multiple stream systems

Author

Airey, S.S. ; Gales, M.J.F.

Author_Institution

Eng. Dept., Cambridge Univ., UK

Volume

1

fYear

2003

fDate

6-10 April 2003

Abstract

There has been interest in the use of classifiers based on the product of experts (PoE). PoEs offer an alternative to the standard mixture of experts (MoE) framework. This paper presents a particular form of PoE, the product of Gaussians (PoG), within an hidden Markov model framework. Training and initialisation procedures are described for this PoG system. In addition, the relationship of PoG to standard multiple stream systems is explored. The PoG system performance is examined on the SwitchBoard task and is compared to standard Gaussian mixture systems and multiple stream systems.

Keywords

Gaussian processes; feature extraction; hidden Markov models; signal classification; speech recognition; PoG system performance; SwitchBoard task; feature vector; hidden Markov model; initialisation procedures; mixture of experts; multiple stream systems; product of Gaussians; product of experts based classifiers; speech recognition; standard Gaussian mixture systems; training procedures; Covariance matrix; Gaussian approximation; Gaussian distribution; Gaussian processes; Hidden Markov models; Speech recognition; System performance; Training data; Vocabulary;

fLanguage

English

Publisher

ieee

Conference_Titel

Acoustics, Speech, and Signal Processing, 2003. Proceedings. (ICASSP '03). 2003 IEEE International Conference on

ISSN

1520-6149

Print_ISBN

0-7803-7663-3

Type

conf

DOI

10.1109/ICASSP.2003.1198913

Filename

1198913