Multilingually Trained Bottleneck Features in Spoken Language Recognition

Czech title

Vícejazyčně trénované parametry založené na úzkém hrdle neuronových sítí pro rozpoznávání mluveného jazyka

Type

journal article

Language

English

Authors

Fér Radek, Ing.
Matějka Pavel, Ing., Ph.D. (DCGM)
Grézl František, Ing., Ph.D. (DCGM)
Plchot Oldřich, Ing., Ph.D. (DCGM)
Veselý Karel, Ing., Ph.D. (DCGM)
Černocký Jan, prof. Dr. Ing. (DCGM)

URL

Keywords

Multilingual training, Bottleneck features, Spoken language recognition

Abstract

Multilingual training of neural networks has proven to be simple yet effective way to deal with multilingual training corpora. It allows to use several resources to jointly train a language independent representation of features, which can be encoded into low-dimensional feature set by embedding narrow bottleneck layer to the network. In this paper, we analyze such features on the task of spoken language recognition (SLR), focusing on practical aspects of training bottleneck networks and analyzing their integration in SLR. By comparing properties of mono and multilingual features we show the suitability of multilingual training for SLR. The state-of-the-art performance of these features is demonstrated on the NIST LRE09 database.

Annotation

Multilingual training of neural networks has proven to be simple yet effective way to deal with multilingual training corpora. It allows to use several resources to jointly train a language independent representation of features, which can be encoded into low-dimensional feature set by embedding narrow bottleneck layer to the network. In this paper, we analyze such features on the task of spoken language recognition (SLR), focusing on practical aspects of training bottleneck networks and analyzing their integration in SLR. By comparing properties of mono and multilingual features we show the suitability of multilingual training for SLR. The state-of-the-art performance of these features is demonstrated on the NIST LRE09 database.

Published

2017

Pages

252–267

Journal

COMPUTER SPEECH AND LANGUAGE, vol. 2017, no. 46, ISSN 0885-2308

DOI

10.1016/j.csl.2017.06.008

UT WoS

000407609600015

EID Scopus

2-s2.0-85023638945

BibTeX

@article{BUT144471,
  author="Radek {Fér} and Pavel {Matějka} and František {Grézl} and Oldřich {Plchot} and Karel {Veselý} and Jan {Černocký}",
  title="Multilingually Trained Bottleneck Features in Spoken Language Recognition",
  journal="COMPUTER SPEECH AND LANGUAGE",
  year="2017",
  volume="2017",
  number="46",
  pages="252--267",
  doi="10.1016/j.csl.2017.06.008",
  issn="0885-2308",
  url="http://www.sciencedirect.com/science/article/pii/S0885230816302947"
}