Conference paper

GRÉZL František, EGOROVA Ekaterina and KARAFIÁT Martin. Further Investigation into Multilingual Training and Adaptation of Stacked Bottle-neck Neural Network Structure. In: Proceedings of 2014 Spoken Language Technology Workshop. South Lake Tahoe, Nevada: IEEE Signal Processing Society, 2014, pp. 48-53. ISBN 978-1-4799-7129-9.
Publication language:english
Original title:Further Investigation into Multilingual Training and Adaptation of Stacked Bottle-neck Neural Network Structure
Title (cs):Pokračující výzkum multilingválního trénování a adaptace neuronových sítí se strukturou stackovaných úzkých vrstev
Pages:48-53
Proceedings:Proceedings of 2014 Spoken Language Technology Workshop
Conference:Spoken Language Technology Workshop (SLT 2014)
Place:South Lake Tahoe, Nevada, US
Year:2014
ISBN:978-1-4799-7129-9
Publisher:IEEE Signal Processing Society
URL:http://www.fit.vutbr.cz/research/groups/speech/publi/2014/grezl_slt2014_0000048.pdf [PDF]
Files: 
+Type Name Title Size Last modified
icongrezl_slt2014_0000048.pdf104 KB2017-03-01 19:02:37
^ Select all
With selected:
Keywords
multilingual training, neural networks, stacked bottle-neck, neural network adaptation
Annotation
This article is about further investigation into multilingual training and adaptation of stacked Bottle-neck Neural Network Structure.
Abstract
Multilingual training of neural networks for ASR is widely studied these days. It has been shown that languages with little training data can benefit largely from multilingual resources. We have evaluated possible ways of adaptation of multilingual stacked bottle-neck hierarchy to target domain. This paper extends our latest work and focuses on the impact certain aspects have on the performance of an adapted neural network feature extractor. First, the performance of adapted multilingual networks preliminarily trained on different languages is studied. Next, the effect of different target units - phonemes vs. triphone states - used for multilingual NN training is evaluated. Then the impact of an increasing number of languages used for multilingual NN training is investigated. Here the condition of constant amount of data is added to separately control the influence of larger language variability and larger amount of data. The effect of adding languages from a different domain is also evaluated. Finally a study is performed where a language with the phonetic structure similar to the target’s one is added to multilingual training data.
BibTeX:
@INPROCEEDINGS{
   author = {Franti{\v{s}}ek Gr{\'{e}}zl and Ekaterina Egorova and Martin
	Karafi{\'{a}}t},
   title = {Further Investigation into Multilingual Training and
	Adaptation of Stacked Bottle-neck Neural Network Structure},
   pages = {48--53},
   booktitle = {Proceedings of 2014 Spoken Language Technology Workshop},
   year = {2014},
   location = {South Lake Tahoe, Nevada, US},
   publisher = {IEEE Signal Processing Society},
   ISBN = {978-1-4799-7129-9},
   language = {english},
   url = {http://www.fit.vutbr.cz/research/view_pub.php?id=10798}
}

Your IPv4 address: 54.225.41.203
Switch to IPv6 connection

DNSSEC [dnssec]