Conference paper

VESELÝ Karel, KARAFIÁT Martin, GRÉZL František, JANDA Miloš and EGOROVA Ekaterina. The Language-Independent Bottleneck Features. In: Proceedings of IEEE 2012 Workshop on Spoken Language Technology. Miami: IEEE Signal Processing Society, 2012, pp. 336-341. ISBN 978-1-4673-5124-9.
Publication language:english
Original title:The Language-Independent Bottleneck Features
Title (cs):Bottleneck příznaky nezávislé na jazyce
Pages:336-341
Proceedings:Proceedings of IEEE 2012 Workshop on Spoken Language Technology
Conference:IEEE 2012 Workshop on Spoken Language Technology
Place:Miami, US
Year:2012
ISBN:978-1-4673-5124-9
Publisher:IEEE Signal Processing Society
URL:http://www.fit.vutbr.cz/research/groups/speech/publi/2012/vesely_slt2012_0000336.pdf [PDF]
Keywords
Language-Independent Bottleneck Features, Multilingual Neural Network

Annotation
The paper is about language-independent bottleneck features, which are generated by Multi-lingual Neural Network. This leads to features which are not biased towards any of the source languages, making the features effectively language independent.
Abstract
In this paper we present novel language-independent bottleneck (BN) feature extraction framework. In our experiments we have used Multilingual Artificial Neural Network (ANN), where each language is modelled by separate output layer, while all the hidden layers jointly model the variability of all the source languages. The key idea is that the entire
ANN is trained on all the languages simultaneously, thus the BN-features are not biased towards any of the languages. Exactly for this reason, the final BN-features are considered as language independent.

In the experiments with GlobalPhone database, we show that the Multilingual BN-features consistently outperform the Monolingual BN-features. Also, the cross-lingual generalisation is evaluated, where we train on 5 source languages and test on 3 other languages. The results show that the ANN can produce very good BN-features even for unseen languages. In some cases even better than if we would train the ANN on the target language only.
BibTeX:
@INPROCEEDINGS{
   author = {Karel Vesel{\'{y}} and Martin Karafi{\'{a}}t and
	Franti{\v{s}}ek Gr{\'{e}}zl and Milo{\v{s}} Janda and
	Ekaterina Egorova},
   title = {The Language-Independent Bottleneck Features},
   pages = {336--341},
   booktitle = {Proceedings of IEEE 2012 Workshop on Spoken Language
	Technology},
   year = {2012},
   location = {Miami, US},
   publisher = {IEEE Signal Processing Society},
   ISBN = {978-1-4673-5124-9},
   language = {english},
   url = {http://www.fit.vutbr.cz/research/view_pub.php.en.iso-8859-2?id=10100}
}

Your IPv4 address: 107.22.126.144
Switch to IPv6 connection

DNSSEC [dnssec]