Conference paper

 
Matějka, P., Černocký, J.: Feature gaussianization in speech recognition, In: Proc. 12th International scientific conference Radioelektronika 2002, Bratislava, SK, STUBA, 2002, p. 4, ISBN 80-227-1700-2
Publication language:english
Original title:Feature gaussianization in speech recognition
Title (cs):Gaussianizace dat v rozpoznávání řeči
Pages:4
Proceedings:Proc. 12th International scientific conference Radioelektronika 2002
Conference:12th International Czech-Slovak Scientific conference Radioelektronika 2002
Place:Bratislava, SK
Year:2002
ISBN:80-227-1700-2
Publisher:Slovak University of Technology in Bratislava
URL:http://www.fit.vutbr.cz/~matejkap/publi/2002/RA2002.pdf [PDF]
Keywords
speech recognition, Gaussianization, non-linear transform
Annotation
Feature gaussianization in speech recognition
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.
BibTeX:
@INPROCEEDINGS{
   author = {Pavel Matějka and Jan Černocký},
   title = {Feature gaussianization in speech recognition},
   pages = {4},
   booktitle = {Proc. 12th International scientific conference
	Radioelektronika 2002},
   year = {2002},
   location = {Bratislava, SK},
   publisher = {Slovak University of Technology in Bratislava},
   ISBN = {80-227-1700-2},
   language = {english},
   url = {http://www.fit.vutbr.cz/research/view_pub.php?id=7082}
}