14thAnnual Conference of the International Speech Communication Association

Lyon, France
August 25-29, 2013

Automatic Regularization of Cross-Entropy Cost for Speaker Recognition Fusion

Ville Hautamäki (1), Kong Aik Lee (2), David A. van Leeuwen (3), R. Saeidi (3), Anthony Larcher (2), Tomi Kinnunen (1), Taufiq Hasan (4), Seyed Omid Sadjadi (4), Gang Liu (4), Hynek Bořil (4), John H. L. Hansen (4), Benoit Fauve (5)

(1) University of Eastern Finland, Finland
(2) A*STAR, Singapore
(3) Radboud Universiteit Nijmegen, The Netherlands
(4) University of Texas at Dallas, USA
(5) ValidSoft Ltd., UK

In this paper we study automatic regularization techniques for the fusion of automatic speaker recognition systems. Parameter regularization could dramatically reduce the fusion training time. In addition, there will not be any need for splitting the development set into different folds for cross- validation. We utilize majorization-minimization approach to automatic ridge regression learning and design a similar way to learn LASSO regularization parameter automatically. By experiments we show improvement in using automatic regularization.

Full Paper

Bibliographic reference.  Hautamäki, Ville / Lee, Kong Aik / Leeuwen, David A. van / Saeidi, R. / Larcher, Anthony / Kinnunen, Tomi / Hasan, Taufiq / Sadjadi, Seyed Omid / Liu, Gang / Bořil, Hynek / Hansen, John H. L. / Fauve, Benoit (2013): "Automatic regularization of cross-entropy cost for speaker recognition fusion", In INTERSPEECH-2013, 1609-1613.