Publication Details

Speaker embeddings by modeling channel-wise correlations

STAFYLAKIS Themos, ROHDIN Johan A. and BURGET Lukáš. Speaker embeddings by modeling channel-wise correlations. In: Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH. Brno: International Speech Communication Association, 2021, pp. 501-505. ISSN 1990-9772. Available from: https://www.isca-speech.org/archive/interspeech_2021/stafylakis21_interspeech.html
Czech title
Embeddingy mluvčích získané pomocí modelování korelací mezi kanály
Type
conference paper
Language
english
Authors
Stafylakis Themos (OMILIA)
Rohdin Johan A., Dr. (DCGM FIT BUT)
Burget Lukáš, doc. Ing., Ph.D. (DCGM FIT BUT)
URL
Keywords

speaker recognition, style-transfer, deep learning

Abstract

Speaker embeddings extracted with deep 2D convolutional neural networks are typically modeled as projections of first and second order statistics of channel-frequency pairs onto a linear layer, using either average or attentive pooling along the time axis. In this paper we examine an alternative pooling method, where pairwise correlations between channels for given frequencies are used as statistics. The method is inspired by style-transfer methods in computer vision, where the style of an image, modeled by the matrix of channel-wise correlations, is transferred to another image, in order to produce a new image having the style of the first and the content of the second. By drawing analogies between image style and speaker characteristics, and between image content and phonetic sequence, we explore the use of such channel-wise correlations features to train a ResNet architecture in an end-to-end fashion. Our experiments on VoxCeleb demonstrate the effectiveness of the proposed pooling method in speaker recognition.

Published
2021
Pages
501-505
Journal
Proceedings of Interspeech - on-line, vol. 2021, no. 8, ISSN 1990-9772
Proceedings
Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH
Conference
Interspeech Conference, Brno, CZ
Publisher
International Speech Communication Association
Place
Brno, CZ
DOI
UT WoS
000841879500101
EID Scopus
BibTeX
@INPROCEEDINGS{FITPUB12596,
   author = "Themos Stafylakis and A. Johan Rohdin and Luk\'{a}\v{s} Burget",
   title = "Speaker embeddings by modeling channel-wise correlations",
   pages = "501--505",
   booktitle = "Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH",
   journal = "Proceedings of Interspeech - on-line",
   volume = 2021,
   number = 8,
   year = 2021,
   location = "Brno, CZ",
   publisher = "International Speech Communication Association",
   ISSN = "1990-9772",
   doi = "10.21437/Interspeech.2021-1442",
   language = "english",
   url = "https://www.fit.vut.cz/research/publication/12596"
}
Back to top