Publication Details
Fast variational Bayes for heavy-tailed PLDA applied to i-vectors and x-vectors
Brummer Johan Nikolaas Langenhoven, Dr. (Phonexia)
García-Romero Daniel (JHU)
Snyder David (JHU)
Burget Lukáš, doc. Ing., Ph.D. (DCGM FIT BUT)
peaker recognition, variational Bayes, heavytailed PLDA
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 SRE10, SRE16 and SITW.
@INPROCEEDINGS{FITPUB11837,
author = "Anna Silnova and Langenhoven Nikolaas Johan Brummer and Daniel Garc\'{i}a-Romero and David Snyder and Luk\'{a}\v{s} Burget",
title = "Fast variational Bayes for heavy-tailed PLDA applied to i-vectors and x-vectors",
pages = "72--76",
booktitle = "Proceedings of Interspeech 2018",
journal = "Proceedings of Interspeech - on-line",
volume = 2018,
number = 9,
year = 2018,
location = "Hyderabad, IN",
publisher = "International Speech Communication Association",
ISSN = "1990-9772",
doi = "10.21437/Interspeech.2018-2128",
language = "english",
url = "https://www.fit.vut.cz/research/publication/11837"
}