Publication Details

Analyzing speaker verification embedding extractors and back-ends under language and channel mismatch

SILNOVA Anna, STAFYLAKIS Themos, MOŠNER Ladislav, PLCHOT Oldřich, ROHDIN Johan A., MATĚJKA Pavel, BURGET Lukáš, GLEMBEK Ondřej and BRUMMER Johan Nikolaas Langenhoven. Analyzing speaker verification embedding extractors and back-ends under language and channel mismatch. In: Proceedings of The Speaker and Language Recognition Workshop (Odyssey 2022). Beijing: International Speech Communication Association, 2022, pp. 9-16. Available from: https://www.isca-speech.org/archive/pdfs/odyssey_2022/silnova22_odyssey.pdf
Czech title
Analýza extraktorů embeddingů a back-endů pro ověřování mluvčího v případech nekoherence přenosového kanálu a jazyka
Type
conference paper
Language
english
Authors
Silnova Anna, MSc., Ph.D. (DCGM FIT BUT)
Stafylakis Themos (OMILIA)
Mošner Ladislav, Ing. (DCGM FIT BUT)
Plchot Oldřich, Ing., Ph.D. (DCGM FIT BUT)
Rohdin Johan A., Dr. (DCGM FIT BUT)
Matějka Pavel, Ing., Ph.D. (DCGM FIT BUT)
Burget Lukáš, doc. Ing., Ph.D. (DCGM FIT BUT)
Glembek Ondřej, Ing., Ph.D. (DCGM FIT BUT)
Brummer Johan Nikolaas Langenhoven, Dr. (Phonexia)
URL
Keywords

speaker, verification, embedding

Abstract

In this paper, we analyze the behavior and performance of speaker embeddings and the back-end scoring model under domain and language mismatch. We present our findings regarding ResNet-based speaker embedding architectures and show that reduced temporal stride yields improved performance. We then consider a PLDA back-end and show how a combination of small speaker subspace, language-dependent PLDA mixture, and nuisance-attribute projection can have a drastic impact on the performance of the system. Besides, we present an efficient way of scoring and fusing class posterior logit vectors recently shown to perform well on speaker verification task. The experiments are performed using the NIST SRE 2021 setup.

Published
2022
Pages
9-16
Proceedings
Proceedings of The Speaker and Language Recognition Workshop (Odyssey 2022)
Conference
Odyssey 2022: The Speaker and Language Recognition Workshop, Beijing, CN
Publisher
International Speech Communication Association
Place
Beijing, CN
DOI
10.21437/Odyssey.2022-2
BibTeX 
@INPROCEEDINGS{FITPUB12834,
   author = "Anna Silnova and Themos Stafylakis and Ladislav Mo\v{s}ner and Old\v{r}ich Plchot and A. Johan Rohdin and Pavel Mat\v{e}jka and Luk\'{a}\v{s} Burget and Ond\v{r}ej Glembek and Langenhoven Nikolaas Johan Brummer",
   title = "Analyzing speaker verification embedding extractors and back-ends under language and channel mismatch",
   pages = "9--16",
   booktitle = "Proceedings of The Speaker and Language Recognition Workshop (Odyssey 2022)",
   year = 2022,
   location = "Beijing, CN",
   publisher = "International Speech Communication Association",
   doi = "10.21437/Odyssey.2022-2",
   language = "english",
   url = "https://www.fit.vut.cz/research/publication/12834"
}
Back to top