Publication Details

Unscented Transform For Ivector-based Noisy Speaker Recognition

MARTÍNEZ GONZÁLEZ, D.; BURGET, L.; STAFYLAKIS, T.; LEI, Y.; KENNY, P.; LLEIDA, E. Unscented Transform For Ivector-based Noisy Speaker Recognition. In Proceedings of ICASSP 2014. Florencie: IEEE Signal Processing Society, 2014. p. 4070-4074. ISBN: 978-1-4799-2892-7.
Czech title
Unscentovaná transformace pro rozpoznávání řečníka v šumu založené na i-vektorech
Type
conference paper
Language
English
Authors
Martínez González David (FIT)
Burget Lukáš, doc. Ing., Ph.D. (DCGM)
Stafylakis Themos
Lei Yun
Kenny Patrick
LLeida Eduardo (FIT)
URL
Keywords

Noise Robust Speaker Recognition, Unscented Transform, Vector Taylor Series, iVector

Abstract

This article is about new version of an unscented transform for Ivector-based noisy speaker recognition.

Annotation

Recently, a new version of the iVector modelling has been proposed for noise robust speaker recognition, where the nonlinear function that relates clean and noisy cepstral coefficients is approximated by a first order vector Taylor series (VTS). In this paper, it is proposed to substitute the first order VTS by an unscented transform, where unlike VTS, the nonlinear function is not applied over the clean model parameters directly, but over a set of sampled points. The resulting points in the transformed space are then used to calculate the model parameters. For very low signal-to-noise ratio improvements in equal error rate of about 7% for a clean backend and of 14.50% for a multistyle backend are obtained.

Published
2014
Pages
4070–4074
Proceedings
Proceedings of ICASSP 2014
ISBN
978-1-4799-2892-7
Publisher
IEEE Signal Processing Society
Place
Florencie
DOI
10.1109/ICASSP.2014.6854361
UT WoS
000343655304013
EID Scopus
2-s2.0-84905284075
BibTeX 
@inproceedings{BUT111555,
  author="David {Martínez González} and Lukáš {Burget} and Themos {Stafylakis} and Yun {Lei} and Patrick {Kenny} and Eduardo {LLeida}",
  title="Unscented Transform For Ivector-based Noisy Speaker Recognition",
  booktitle="Proceedings of ICASSP 2014",
  year="2014",
  pages="4070--4074",
  publisher="IEEE Signal Processing Society",
  address="Florencie",
  doi="10.1109/ICASSP.2014.6854361",
  isbn="978-1-4799-2892-7",
  url="https://www.fit.vut.cz/research/publication/10573/"
}
Back to top