Two approaches are presented for non-linear PLDA to be used in speaker verification. In NIST 2010 speaker recognition evaluation (SRE) tests under DET-5 conditions, the two methods and particularly their combination provided significant improvements in equal error rates and minDCF values over a standard PLDA scheme. The proposed schemes were also applied within a speaker verification system that employs DNN-based sufficient statistics calculation resulting in a 45% reduction in minDCF relative to a conventional GMM based system.
Cite as: Novoselov, S., Pekhovsky, T., Kudashev, O., Mendelev, V.S., Prudnikov, A. (2015) Non-linear PLDA for i-vector speaker verification. Proc. Interspeech 2015, 214-218, doi: 10.21437/Interspeech.2015-87
@inproceedings{novoselov15_interspeech,
author={Sergey Novoselov and Timur Pekhovsky and Oleg Kudashev and Valentin S. Mendelev and Alexey Prudnikov},
title={{Non-linear PLDA for i-vector speaker verification}},
year=2015,
booktitle={Proc. Interspeech 2015},
pages={214--218},
doi={10.21437/Interspeech.2015-87}
}