Publication Details

Analysis and Description of ABC Submission to NIST SRE 2016

PLCHOT, O.; MATĚJKA, P.; SILNOVA, A.; NOVOTNÝ, O.; DIEZ SÁNCHEZ, M.; ROHDIN, J.; GLEMBEK, O.; BRÜMMER, N.; SWART, A.; PRIETO, J.; GARCIA PERERA, L.; BUERA, L.; KENNY, P.; ALAM, J.; BHATTACHARYA, G. Analysis and Description of ABC Submission to NIST SRE 2016. In Proceedings of Interspeech 2017. Proceedings of Interspeech. Stockholm: International Speech Communication Association, 2017. p. 1348-1352. ISSN: 1990-9772.
Czech title
Analýza a popis ABC systému pro NIST SRE 2016
Type
conference paper
Language
English
Authors
Plchot Oldřich, Ing., Ph.D. (DCGM)
Matějka Pavel, Ing., Ph.D. (DCGM)
Silnova Anna, M.Sc., Ph.D. (DCGM)
Novotný Ondřej, Ing., Ph.D.
Diez Sánchez Mireia, M.Sc., Ph.D. (DCGM)
Rohdin Johan Andréas, M.Sc., Ph.D. (DCGM)
Glembek Ondřej, Ing., Ph.D.
Brümmer Niko
Swart Albert du Preez
Prieto Jesús (FIT)
Garcia Perera Leibny Paola (FIT)
Buera Luis (FIT)
Kenny Patrick
Alam Jahangir
Bhattacharya Gautam (FIT)
URL
Keywords

speaker recognition, i-vector, DNN, fusion

Abstract

This article is about the analysis and description of ABC Submission to NIST SRE 2016.We have presented various sytems of the ABC team that are designed to cope with dataset mismatch and non-English data. We have presented and compared several fusion and calibration strategies and we have uncovered and discussed problems brought by SRE16.

Annotation

We present a condensed description and analysis of the joint submission for NIST SRE 2016, by Agnitio, BUT and CRIM (ABC). We concentrate on challenges that arose during development and we analyze the results obtained on the evaluation data and on our development sets. We show that testing on mismatched, non-English and short duration data introduced in NIST SRE 2016 is a difficult problem for current state-of-theart systems. Testing on this data brought back the issue of score normalization and it also revealed that the bottleneck features (BN), which are superior when used for telephone English, are lacking in performance against the standard acoustic features like Mel Frequency Cepstral Coefficients (MFCCs). We offer ABCs insights, findings and suggestions for building a robust system suitable for mismatched, non-English and relatively noisy data such as those in NIST SRE 2016.

Published
2017
Pages
1348–1352
Journal
Proceedings of Interspeech, vol. 2017, no. 08, ISSN 1990-9772
Proceedings
Proceedings of Interspeech 2017
Publisher
International Speech Communication Association
Place
Stockholm
DOI
10.21437/Interspeech.2017-1498
UT WoS
000457505000280
EID Scopus
2-s2.0-85039148308
BibTeX 
@inproceedings{BUT144490,
  author="Oldřich {Plchot} and Pavel {Matějka} and Anna {Silnova} and Ondřej {Novotný} and Mireia {Diez Sánchez} and Johan Andréas {Rohdin} and Ondřej {Glembek} and Niko {Brümmer} and Albert du Preez {Swart} and Jesús {Prieto} and Leibny Paola {Garcia Perera} and Luis {Buera} and Patrick {Kenny} and Jahangir {Alam} and Gautam {Bhattacharya}",
  title="Analysis and Description of ABC Submission to NIST SRE 2016",
  booktitle="Proceedings of Interspeech 2017",
  year="2017",
  journal="Proceedings of Interspeech",
  volume="2017",
  number="08",
  pages="1348--1352",
  publisher="International Speech Communication Association",
  address="Stockholm",
  doi="10.21437/Interspeech.2017-1498",
  issn="1990-9772",
  url="http://www.isca-speech.org/archive/Interspeech_2017/pdfs/1498.PDF"
}
Back to top