Publication Details

Further Investigation into Multilingual Training and Adaptation of Stacked Bottle-neck Neural Network Structure

GRÉZL, F.; EGOROVA, E.; KARAFIÁT, M. 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. p. 48-53. ISBN: 978-1-4799-7129-9.
Czech title
Pokračující výzkum multilingválního trénování a adaptace neuronových sítí se strukturou stackovaných úzkých vrstev
Type
conference paper
Language
English
Authors
Grézl František, Ing., Ph.D. (DCGM)
Egorova Ekaterina, Ing., Ph.D.
Karafiát Martin, Ing., Ph.D. (DCGM)
URL
Keywords

multilingual training, neural networks, stacked bottle-neck, neural network adaptation

Abstract

This article is about further investigation into multilingual training and adaptation of stacked Bottle-neck Neural Network Structure.

Annotation

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.

Published
2014
Pages
48–53
Proceedings
Proceedings of 2014 Spoken Language Technology Workshop
ISBN
978-1-4799-7129-9
Publisher
IEEE Signal Processing Society
Place
South Lake Tahoe, Nevada
DOI
10.1109/SLT.2014.7078548
UT WoS
000380375100008
EID Scopus
2-s2.0-84946685938
BibTeX 
@inproceedings{BUT111502,
  author="František {Grézl} and Ekaterina {Egorova} and Martin {Karafiát}",
  title="Further Investigation into Multilingual Training and Adaptation of Stacked Bottle-neck Neural Network Structure",
  booktitle="Proceedings of 2014 Spoken Language Technology Workshop",
  year="2014",
  pages="48--53",
  publisher="IEEE Signal Processing Society",
  address="South Lake Tahoe, Nevada",
  doi="10.1109/SLT.2014.7078548",
  isbn="978-1-4799-7129-9",
  url="https://www.fit.vut.cz/research/publication/10798/"
}
Back to top