Publication result detail

Feature Gaussianization for Speech Recognition

MATĚJKA, P., ČERNOCKÝ, J.

Original Title

Feature Gaussianization for Speech Recognition

English Title

Feature Gaussianization for Speech Recognition

Type

Paper in proceedings (conference paper)

Original Abstract

In Hidden Markov models, speech data are modeled by Gaussian distributions. In this paper, we propose to Gaussianize the features to better fit to this modeling. A distribution of the data is estimated and a transform function is derived. We test three methods of the transform estimation (global, speaker based, frame based) and report results on the SPINE 2000 task with Sphinx recognizer. We conclude that the proposed method is a cheap way to increase the recognition accuracy.

English abstract

In Hidden Markov models, speech data are modeled by Gaussian distributions. In this paper, we propose to Gaussianize the features to better fit to this modeling. A distribution of the data is estimated and a transform function is derived. We test three methods of the transform estimation (global, speaker based, frame based) and report results on the SPINE 2000 task with Sphinx recognizer. We conclude that the proposed method is a cheap way to increase the recognition accuracy.

Key words in English

speaker speach recognition gaussianization HMM

Authors

MATĚJKA, P., ČERNOCKÝ, J.

Released

14.05.2002

Publisher

Slovak University of Technology in Bratislava

Location

Bratislava,Slovak Republic

ISBN

80-227-1700-2

Book

Conference Proceedings

Pages from

93

Pages count

4

BibTex

@inproceedings{BUT5411,
  author="Pavel {Matějka} and Jan {Černocký}",
  title="Feature Gaussianization for Speech Recognition",
  booktitle="Conference Proceedings",
  year="2002",
  pages="4",
  publisher="Slovak University of Technology in Bratislava",
  address="Bratislava,Slovak Republic",
  isbn="80-227-1700-2"
}