Publication Details

Variance-Spectra based Normalization for I-vector Standard and Probabilistic Linear Discriminant Analysis

BOUSQUET Pierre-Michel, LARCHER Anthony, MATROUF Driss, BONASTRE Jean-Francois and PLCHOT Oldřich. Variance-Spectra based Normalization for I-vector Standard and Probabilistic Linear Discriminant Analysis. In: Proceedings of Odyssey 2012: The Speaker and Language Recognition Workshop. Singapur: International Speech Communication Association, 2012, pp. 157-164. ISBN 978-981-07-3093-2. Available from: http://www.fit.vutbr.cz/research/groups/speech/publi/2012/bousquet_odyssey2012_157-164-09.pdf
Czech title
Normalizace I-vektorů na základě variance spektra pro standardní a pravděpodobnostní lineární diskriminační analýzu
Type
conference paper
Language
english
Authors
Bousquet Pierre-Michel (UAPV)
Larcher Anthony, Dr. (LMNU)
Matrouf Driss (UAPV)
Bonastre Jean-Francois (UAPV)
Plchot Oldřich, Ing., Ph.D. (FIT BUT)
URL
Keywords

i-vectors, probabilistic linear discriminant analysis, speaker recognition

Abstract

This paper is on various i-vector normalizations for speaker recognition using standard and probabilistic Linear Discriminant Analysis (LDA and PLDA)

Annotation

I-vector extraction and Probabilistic Linear Discriminant Analysis (PLDA) has become the state-of-the-art configuration for speaker verification. Recently, Gaussian-PLDA has been improved by a preliminary length normalization of i-vectors. This normalization, known to increase the Gaussianity of the i-vector distribution, also improves performance of systems based on standard Linear Discriminant Analysis (LDA) and "two-covariance model" scoring. We propose in this paper to replace length normalization by two new techniques based on total, between- and within-speaker variance spectra 1. These "spectral" techniques both normalize the i-vectors length for Gaussianity, but the first adapts the i-vectors representation to a speaker recognition system based on LDA and two-covariance scoring when the second adapts it to a Gaussian-PLDA model. Significant performance improvements are demonstrated on the male and female telephone portion of NIST SRE 2010. Index Terms: i-vectors, probabilistic linear discriminant analysis, speaker recognition.

Published
2012
Pages
157-164
Proceedings
Proceedings of Odyssey 2012: The Speaker and Language Recognition Workshop
Conference
Odyssey 2012: The Speaker and Language Recognition Workshop, Singapur, SG
ISBN
978-981-07-3093-2
Publisher
International Speech Communication Association
Place
Singapur, SG
EID Scopus
2-s2.0-85073229756
BibTeX 
@INPROCEEDINGS{FITPUB10054,
   author = "Pierre-Michel Bousquet and Anthony Larcher and Driss Matrouf and Jean-Francois Bonastre and Old\v{r}ich Plchot",
   title = "Variance-Spectra based Normalization for I-vector Standard and Probabilistic Linear Discriminant Analysis",
   pages = "157--164",
   booktitle = "Proceedings of Odyssey 2012: The Speaker and Language Recognition Workshop",
   year = 2012,
   location = "Singapur, SG",
   publisher = "International Speech Communication Association",
   ISBN = "978-981-07-3093-2",
   language = "english",
   url = "https://www.fit.vut.cz/research/publication/10054"
}
Back to top