Publication Details
BAT System Description for NIST LRE 2015
Matějka Pavel, Ing., Ph.D. (DCGM FIT BUT)
Fér Radek, Ing. (DCGM FIT BUT)
Glembek Ondřej, Ing., Ph.D. (DCGM FIT BUT)
Novotný Ondřej, Ing., Ph.D. (DCGM FIT BUT)
Pešán Jan, Ing. (DCGM FIT BUT)
Veselý Karel, Ing., Ph.D. (DCGM FIT BUT)
Ondel Yang Lucas Antoine Francois, Mgr., Ph.D. (DCGM FIT BUT)
Karafiát Martin, Ing., Ph.D. (DCGM FIT BUT)
Grézl František, Ing., Ph.D. (DCGM FIT BUT)
Kesiraju Santosh (DCGM FIT BUT)
Burget Lukáš, doc. Ing., Ph.D. (DCGM FIT BUT)
Brummer Johan Nikolaas Langenhoven (Agnitio)
Swart Albert du Preez (Agnitio)
Cumani Sandro (POLITO)
Mallidi Sri Harish (AmazonCom)
Li Ruizhi (JHU)
BAT System Description, NIST LRE
In this paper, we summarize our efforts in the NIST Language Recognition (LRE) 2015 Evaluations which resulted in systems providing very competitive performance. We provide both the descriptions and the analysis of the systems that we included in our submission. We start by detailed description of the datasets that we used for training and development, and we follow by describing the models and methods that were used to produce the final scores. These include the front-end (i.e., the voice activity detection and feature extraction), the back-end (i.e., the final classifier), and the calibration and fusion stages. Apart from the techniques commonly used in the field (such as i-vectors, DNN bottle-Neck features, NN classifiers, Gaussian Back-ends, etc.), we present less-common methods, such as Sequence Summarizing Neural Networks (SSNN), and Automatic Unit Discovery. We present the performance of the systems both on the Fixed condition (where participants are required to use predefined data sets only), and the Open condition (where participants are allowed to use any publicly available resource) of the NIST LRE 2015.
In this work, we have described our efforts in the NIST LRE 2015. The most difficult part of this evaluation was to deal with limited amount of data and the results show that the proper analysis in this direction is necessary. We have built over 20 systems for this evaluation. We have experimented with de-noising NN, automatic unit discovery, different flavors of phonotactic systems, backends, sizes of ivector systems, feature sets, BN features or frame level language classifiers. We used up to 6 systems in the fusion. The performance of our best system reached Cavg of 16.9% on the fixed training data condition and 13.9% (11.9% after post-evaluation analysis) on the open training data condition.
@INPROCEEDINGS{FITPUB11221,
author = "Old\v{r}ich Plchot and Pavel Mat\v{e}jka and Radek F\'{e}r and Ond\v{r}ej Glembek and Ond\v{r}ej Novotn\'{y} and Jan Pe\v{s}\'{a}n and Karel Vesel\'{y} and Francois Antoine Lucas Yang Ondel and Martin Karafi\'{a}t and Franti\v{s}ek Gr\'{e}zl and Santosh Kesiraju and Luk\'{a}\v{s} Burget and Langenhoven Nikolaas Johan Brummer and Preez du Albert Swart and Sandro Cumani and Harish Sri Mallidi and Ruizhi Li",
title = "BAT System Description for NIST LRE 2015",
pages = "166--173",
booktitle = "Proceedings of Odyssey 2016, The Speaker and Language Recognition Workshop",
journal = "Proceedings of Odyssey: The Speaker and Language Recognition Workshop",
volume = 2016,
number = 06,
year = 2016,
location = "Bilbao, ES",
publisher = "International Speech Communication Association",
ISSN = "2312-2846",
doi = "10.21437/Odyssey.2016-24",
language = "english",
url = "https://www.fit.vut.cz/research/publication/11221"
}