|
Abstract
|
|
|---|---|
| The present paper presents the system developed to participate in the 2013 Speaker Recognition Evaluation in Mobile Environments. The aim of the system is to show that selecting an adequate front-end that effectively character-izes the speaker is as important as the selection of the classifier. This compo-nent of the recognition system seems to be relegated to background, as demon-strated by the fact that almost all systems submitted to the 2013 Speaker Recognition Evaluation (SRE) in Mobile Environment [1] use roughly the same front-end, i.e. MFCC coefficients extracted from the power spectral density of speech as a whole. The characterization of speakers using these coefficients seems to have become the de facto standard in the area of speaker recognition. In order to show the importance of the front-end in the overall recognition sys-tem, we will compare the recognition rates achieved by different SR systems that rely on the same classifier (GMM-UBM) but make use of different feature extraction systems (based on both classical and biometric parameters). The re-sults presented in [1] confirms that a gender dependent biometric parameteriza-tion with a simple recognition system based on GMM-UBM paradigm provides very competitive or even better recognition rates when compared to more com-plex classification systems based on classical features. | |
|
International
|
No |
|
|
|
|
Book Edition
|
|
|
Book Publishing
|
Universidad de Salamanca |
|
ISBN
|
978-84-616-5690-5 |
|
Series
|
|
|
Book title
|
Actas de las VII Jornadas de Reconocimiento Biométrico de Personas |
|
From page
|
28 |
|
To page
|
37 |