Audiovisual Speaker Identity Verification Based on Cross Modal Fusion

Girija Chetty; Michael Wagner

Audiovisual Speaker Identity Verification Based on Cross Modal Fusion

Girija Chetty, Michael Wagner

Information Systems

Research output: Contribution to conference (non-published works) › Paper › peer-review

1 Citation (Scopus)

33 Downloads (Pure)

Abstract

In this paper, we propose the fusion of audio and explicit correlation features for speaker identity verification applications. Experiments performed with the GMM based speaker models with hybrid fusion technique involving late fusion of explicit cross-modal fusion features, with eigen lip and audio MFCC features allow a considerable improvement in EER performance An evaluation of the system performance with different gender specific datasets from controlled VidTIMIT data base and opportunistic UCBN database shows, that is possible to achieve an EER of less than 2% with correlated component hybrid fusion, and improvement of around 22 % over uncorrelated component fusion.

Original language	English
Pages	1-5
Number of pages	5
Publication status	Published - 2007
Event	International Conference on Audio-Visual Speech Processing - Hilvarenbeek, Netherlands Duration: 31 Aug 2007 → 3 Sept 2007

Conference

Conference	International Conference on Audio-Visual Speech Processing
Country/Territory	Netherlands
City	Hilvarenbeek
Period	31/08/07 → 3/09/07

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

Access to Document

av07_P37_chettyLicence: Other

Cite this

@conference{60d44b7087bd4530bb2a4744846c5403,

title = "Audiovisual Speaker Identity Verification Based on Cross Modal Fusion",

abstract = "In this paper, we propose the fusion of audio and explicit correlation features for speaker identity verification applications. Experiments performed with the GMM based speaker models with hybrid fusion technique involving late fusion of explicit cross-modal fusion features, with eigen lip and audio MFCC features allow a considerable improvement in EER performance An evaluation of the system performance with different gender specific datasets from controlled VidTIMIT data base and opportunistic UCBN database shows, that is possible to achieve an EER of less than 2% with correlated component hybrid fusion, and improvement of around 22 % over uncorrelated component fusion.",

keywords = "audiovisual, cross modal association, liveness checking, speaker identity verification",

author = "Girija Chetty and Michael Wagner",

note = "Publisher Copyright: {\textcopyright} 2007 Auditory-Visual Speech Processing 2007, AVSP 2007. All rights reserved.; International Conference on Audio-Visual Speech Processing ; Conference date: 31-08-2007 Through 03-09-2007",

year = "2007",

language = "English",

pages = "1--5",

}

TY - CONF

T1 - Audiovisual Speaker Identity Verification Based on Cross Modal Fusion

AU - Chetty, Girija

AU - Wagner, Michael

PY - 2007

Y1 - 2007

N2 - In this paper, we propose the fusion of audio and explicit correlation features for speaker identity verification applications. Experiments performed with the GMM based speaker models with hybrid fusion technique involving late fusion of explicit cross-modal fusion features, with eigen lip and audio MFCC features allow a considerable improvement in EER performance An evaluation of the system performance with different gender specific datasets from controlled VidTIMIT data base and opportunistic UCBN database shows, that is possible to achieve an EER of less than 2% with correlated component hybrid fusion, and improvement of around 22 % over uncorrelated component fusion.

AB - In this paper, we propose the fusion of audio and explicit correlation features for speaker identity verification applications. Experiments performed with the GMM based speaker models with hybrid fusion technique involving late fusion of explicit cross-modal fusion features, with eigen lip and audio MFCC features allow a considerable improvement in EER performance An evaluation of the system performance with different gender specific datasets from controlled VidTIMIT data base and opportunistic UCBN database shows, that is possible to achieve an EER of less than 2% with correlated component hybrid fusion, and improvement of around 22 % over uncorrelated component fusion.

KW - audiovisual

KW - cross modal association

KW - liveness checking

KW - speaker identity verification

UR - http://www.scopus.com/inward/record.url?scp=85133391648&partnerID=8YFLogxK

M3 - Paper

SP - 1

EP - 5

T2 - International Conference on Audio-Visual Speech Processing

Y2 - 31 August 2007 through 3 September 2007

ER -

Audiovisual Speaker Identity Verification Based on Cross Modal Fusion

Abstract

Conference

UN SDGs

Access to Document

Other files and links

Fingerprint

Cite this