Conference paper

BRUMMER Niko, SILNOVA Anna, BURGET Lukáš and STAFYLAKIS Themos. Gaussian meta-embeddings for efficient scoring of a heavy-tailed PLDA model. In: Proceedings of Odyssey 2018. Les Sables d'Olonne: International Speech Communication Association, 2018, pp. 349-356. ISSN 2312-2846.
Publication language:english
Original title:Gaussian meta-embeddings for efficient scoring of a heavy-tailed PLDA model
Title (cs):Gaussovské meta-embeddingy pro efektivní skórování PLDA modelu s těžkým chvostem.
Pages:349-356
Proceedings:Proceedings of Odyssey 2018
Conference:Odyssey 2018
Place:Les Sables d'Olonne, FR
Year:2018
Journal:Proceedings of Odyssey: The Speaker and Language Recognition Workshop, Vol. 2018, No. 6, 4 Rue des Fauvettes - Lous Tourils, F-66390 BAIXAS, FR
ISSN:2312-2846
DOI:10.21437/Odyssey.2018-49
Publisher:International Speech Communication Association
URL:http://www.fit.vutbr.cz/research/groups/speech/publi/2018/brummer_odyssey2018_51.pdf [PDF]
Keywords
embeddings, machine learning, speaker recognition
Annotation
Embeddings in machine learning are low-dimensional representations of complex input patterns, with the property that simple geometric operations like Euclidean distances and dot products can be used for classification and comparison tasks. We introduce meta-embeddings, which live in more general inner product spaces and which are designed to better propagate uncertainty through the embedding bottleneck. Traditional embeddings are trained to maximize between-class and minimize within-class distances. Meta-embeddings are trained to maximize relevant information throughput. As a proof of concept in speaker recognition, we derive an extractor from the familiar generative Gaussian PLDA model (GPLDA). We show that GPLDA likelihood ratio scores are given by Hilbert space inner products between Gaussian likelihood functions, which we term Gaussian meta-embeddings (GMEs). Meta-embedding extractors can be generatively or discriminatively trained. GMEs extracted by GPLDA have fixed precisions and do not propagate uncertainty. We show that a generalization to heavy-tailed PLDA gives GMEs with variable precisions, which do propagate uncertainty. Experiments on NIST SRE 2010 and 2016 show that the proposed method applied to i-vectors without length normalization is up to 20% more accurate than GPLDA applied to length-normalized i-vectors.
BibTeX:
@INPROCEEDINGS{
   author = {Niko Brummer and Anna Silnova and
	Luk{\'{a}}{\v{s}} Burget and Themos Stafylakis},
   title = {Gaussian meta-embeddings for efficient scoring of
	a heavy-tailed PLDA model},
   pages = {349--356},
   booktitle = {Proceedings of Odyssey 2018},
   journal = {Proceedings of Odyssey: The Speaker and Language Recognition
	Workshop},
   volume = {2018},
   number = {6},
   year = {2018},
   location = {Les Sables d'Olonne, FR},
   publisher = {International Speech Communication Association},
   ISSN = {2312-2846},
   doi = {10.21437/Odyssey.2018-49},
   language = {english},
   url = {http://www.fit.vutbr.cz/research/view_pub.php?id=11790}
}

Your IPv4 address: 54.159.51.118
Switch to IPv6 connection

DNSSEC [dnssec]