Discriminative Classifiers for Speaker Recognition - Couverture souple

Katz, Marcel

 
9783838101910: Discriminative Classifiers for Speaker Recognition
Couverture souple
ISBN 10 : 383810191X ISBN 13 : 9783838101910
Editeur : Südwestdeutscher Verlag für Hochschulschriften, 2009

Synopsis

Due to the growing need for security applications,speaker recognition as the biometric task ofauthenticating a claimant by voice has currentlybecome a focus of interest.In this book we present new approaches to integratediscriminative classifiers like Support VectorMachines (SVMs) and Sparse Kernel Logistic Regression(SKLR) into speaker recognition systems that aretraditionally based on generative classifiers likeGaussian Mixture Models (GMMs).In a first approach for limited training data thediscriminative classifiers are applied directly onfeature vectors from parameterized speech frames andit is shown that both, SVM as well as SKLR outperformtraditional methods.In the second approach a state-of-the-art speakerrecognition system for large amount of training datais designed that combines Gaussian Mixture Modelswith discriminative classifiers.Furthermore, we investigate different featureextraction methods for speaker recognition on largeamount of training data and it is shown that theapplication of fusion schemes that combine thesesubsystems yield a significant improvement of therecognition performance in comparison to theapplication of single subsystems.

Les informations fournies dans la section « Synopsis » peuvent faire référence à une autre édition de ce titre.

Présentation de l'éditeur

Due to the growing need for security applications,speaker recognition as the biometric task ofauthenticating a claimant by voice has currentlybecome a focus of interest.In this book we present new approaches to integratediscriminative classifiers like Support VectorMachines (SVMs) and Sparse Kernel Logistic Regression(SKLR) into speaker recognition systems that aretraditionally based on generative classifiers likeGaussian Mixture Models (GMMs).In a first approach for limited training data thediscriminative classifiers are applied directly onfeature vectors from parameterized speech frames andit is shown that both, SVM as well as SKLR outperformtraditional methods.In the second approach a state-of-the-art speakerrecognition system for large amount of training datais designed that combines Gaussian Mixture Modelswith discriminative classifiers.Furthermore, we investigate different featureextraction methods for speaker recognition on largeamount of training data and it is shown that theapplication of fusion schemes that combine thesesubsystems yield a significant improvement of therecognition performance in comparison to theapplication of single subsystems.

Biographie de l'auteur

Marcel Katz studied Electrical Engineering in Duesseldorf andreceived his PhD for his works in the fields of speech andspeaker recognition in 2008 from the University of Magdeburg,Germany. He successfully participated in several speakerrecognition evaluations and is currently working as a speechrecognition specialist in Cambridge, UK.

Les informations fournies dans la section « A propos du livre » peuvent faire référence à une autre édition de ce titre.

Éditeur
Südwestdeutscher Verlag für Hochschulschriften
Date d'édition
2009
Langue
allemand
ISBN 10 
383810191X
ISBN 13 
9783838101910
Reliure
Broché
Nombre de pages
160
Coordonnées du fabricant
non disponible
Personne responsable
non disponible