Fast variational Bayes for heavy-tailed PLDA applied to i-vectors and x-vectors
Abstract
The standard state-of-the-art backend for text-independent speaker recognizers that use i-vectors or x-vectors, is Gaussian PLDA (G-PLDA), assisted by a Gaussianization step involving length normalization. G-PLDA can be trained with both generative or discriminative methods. It has long been known that heavy-tailed PLDA (HT-PLDA), applied without length normalization, gives similar accuracy, but at considerable extra computational cost. We have recently introduced a fast scoring algorithm for a discriminatively trained HT-PLDA backend. This paper extends that work by introducing a fast, variational Bayes, generative training algorithm. We compare old and new backends, with and without length-normalization, with i-vectors and x-vectors, on SRE'10, SRE'16 and SITW.
Keywords
Cite
@article{arxiv.1803.09153,
title = {Fast variational Bayes for heavy-tailed PLDA applied to i-vectors and x-vectors},
author = {Anna Silnova and Niko Brummer and Daniel Garcia-Romero and David Snyder and Lukas Burget},
journal= {arXiv preprint arXiv:1803.09153},
year = {2018}
}
Comments
Submittted to Interspeech 2018