Publication Details
BUT/Phonexia Bottleneck Feature Extractor
Matějka Pavel, Ing., Ph.D. (DCGM)
Glembek Ondřej, Ing., Ph.D.
Plchot Oldřich, Ing., Ph.D. (DCGM)
Novotný Ondřej, Ing., Ph.D.
Grézl František, Ing., Ph.D. (DCGM)
Schwarz Petr, Ing., Ph.D. (DCGM)
Černocký Jan, prof. Dr. Ing. (DCGM)
bottlneck feature extractor, speech recognition, language recognition
This paper complements the public release of theBUT/Phonexia bottleneck (BN) feature extractor. Startingwith a brief history of Neural Network (NN)-based andBN-based approaches to speech feature extraction, it describesthe structure of the released software. It follows by describingthe three provided NNs: the first two trained on the US EnglishFisher corpus with monophone-state and tied-state targets,and the third network trained in a multi-lingual fashion on17 Babel languages. The NNs were technically trained toclassify acoustic units, however the networks were optimizedwith respect to the language recognition task, which is themain focus of this paper. Nevertheless, it is worth noting thatapart from language recognition, the provided software can beused with any speech-related task. The paper concludes with acomprehensive summary of the results obtained on the NIST2015 and 2017 Language Recognition Evaluations tasks.
@inproceedings{BUT155076,
author="Anna {Silnova} and Pavel {Matějka} and Ondřej {Glembek} and Oldřich {Plchot} and Ondřej {Novotný} and František {Grézl} and Petr {Schwarz} and Jan {Černocký}",
title="BUT/Phonexia Bottleneck Feature Extractor",
booktitle="Proceedings of Odyssey 2018",
year="2018",
journal="Proceedings of Odyssey: The Speaker and Language Recognition Workshop Odyssey 2014, Joensuu, Finland",
volume="2018",
number="6",
pages="283--287",
publisher="International Speech Communication Association",
address="Les Sables d´Olonne",
doi="10.21437/Odyssey.2018-40",
issn="2312-2846",
url="https://www.fit.vut.cz/research/publication/11789/"
}