Comparison of scoring methods used in speaker recognition with Joint Factor Analysis

Authors:
Ondrej Glembek;Lukas Burget;Najim Dehak;Niko Brummer;Patrick Kenny
Affiliations:
Speech@FIT group, Faculty of Information Technology, Brno University of Technology, Czech Republic;Speech@FIT group, Faculty of Information Technology, Brno University of Technology, Czech Republic;Centre de Recherche Informatique de Montréal (CRIM), Montréal, Canada;Agnitio, Stellenbosch, South Africa;Centre de Recherche Informatique de Montréal (CRIM), Montréal, Canada
Venue:
ICASSP '09 Proceedings of the 2009 IEEE International Conference on Acoustics, Speech and Signal Processing
Year:
2009

Citing 0
Cited 6

An overview of text-independent speaker recognition: From features to supervectors

Speech Communication
Towards automatic speaker retrieval for large multimedia archives

Proceedings of the 3rd international workshop on Automated information extraction in media production
Modeling nuisance variabilities with factor analysis for GMM-based audio pattern classification

Computer Speech and Language
Detecting replay attacks from far-field recordings on speaker verification systems

BioID'11 Proceedings of the COST 2101 European conference on Biometrics and ID management
Orthogonal subspace combination based on the joint factor analysis for text-independent speaker recognition

CCBR'12 Proceedings of the 7th Chinese conference on Biometric Recognition
Comparison between supervised and unsupervised learning of probabilistic linear discriminant analysis mixture models for speaker verification

Pattern Recognition Letters

Quantified Score

Hi-index	0.00

Visualization

Abstract

The aim of this paper is to compare different log-likelihood scoring methods, that different sites used in the latest state-of-the-art Joint Factor Analysis (JFA) Speaker Recognition systems. The algorithms use various assumptions and have been derived from various approximations of the objective functions of JFA. We compare the techniques in terms of speed and performance. We show, that approximations of the true log-likelihood ratio (LLR) may lead to significant speedup without any loss in performance.