LOCATION: Auditorium 1, TEL, Ernst-Reuter-Platz 7, 20^th floor

Date/Time: 26.10.2015, 14:15-15:00

SPEAKER: Michael Wagner (TU Berlin)

ABSTRACT:

The human voice uniquely combines physical and behavioural traits of the individual. Large intra-person variance, channel and environmental variation present challenges for speaker verification, while the universality of speech and the ubiquity of the telephone make the voice a very useful biometric for authentication and forensic applications. The field has seem enormous progress in recent years with equal-error rates for clean signal having dropped to below 1% in state-of-the-art systems. Recent advances based on the UBM-GMM framework, including factor analysis, iVectors and PLDA as well as calibration and specific forensic speaker comparison issues, are presented in this talk.

BIO:

Michael Wagner is an Honorary Professor at TU Berlin, Adjunct Professor at the Australian National University and Managing Director of the National Centre for Biometric Studies Pty Ltd.