Improving robustness of speaker recognition to new conditions using unlabeled data
Unsupervised techniques for the adaptation of speaker recognition are important due to the problem of condition mismatch that is prevalent when applying speaker recognition technology to new conditions and the general scarcity of labeled 'in-domain' data. In the recent NIST 2016 Speaker Re...
Guardado en:
Autores principales: | , , , , , , , , , , |
---|---|
Formato: | CONF |
Materias: | |
Acceso en línea: | http://hdl.handle.net/20.500.12110/paper_2308457X_v2017-August_n_p3737_Castan |
Aporte de: |
id |
todo:paper_2308457X_v2017-August_n_p3737_Castan |
---|---|
record_format |
dspace |
spelling |
todo:paper_2308457X_v2017-August_n_p3737_Castan2023-10-03T16:40:54Z Improving robustness of speaker recognition to new conditions using unlabeled data Castan, D. McLaren, M. Ferrer, L. Lawson, A. Lozano-Diez, A. Lacerda F. Strombergsson S. Wlodarczak M. Heldner M. Gustafson J. House D. NIST SRE16 Score Calibration Score Normalization Trial-based Calibration Calibration Speech communication Acoustic conditions Calibration parameters NIST SRE16 Score normalization Speaker clustering Speaker recognition Speaker recognition evaluations Unsupervised techniques Speech recognition Unsupervised techniques for the adaptation of speaker recognition are important due to the problem of condition mismatch that is prevalent when applying speaker recognition technology to new conditions and the general scarcity of labeled 'in-domain' data. In the recent NIST 2016 Speaker Recognition Evaluation (SRE), symmetric score normalization (Snorm) and calibration using unlabeled in-domain data were shown to be beneficial. Because calibration requires speaker labels for training, speaker-clustering techniques were used to generate pseudo-speakers for learning calibration parameters in those cases where only unlabeled in-domain data was available. These methods performed well in the SRE16. It is unclear, however, whether those techniques generalize well to other data sources. In this work, we benchmark these approaches on several distinctly different databases, after we describe our SRI-CON-UAM team system submission for the NIST 2016 SRE. Our analysis shows that while the benefit of S-norm is also observed across other datasets, applying speaker-clustered calibration provides considerably greater benefit to the system in the context of new acoustic conditions. Copyright © 2017 ISCA. CONF info:eu-repo/semantics/openAccess http://creativecommons.org/licenses/by/2.5/ar http://hdl.handle.net/20.500.12110/paper_2308457X_v2017-August_n_p3737_Castan |
institution |
Universidad de Buenos Aires |
institution_str |
I-28 |
repository_str |
R-134 |
collection |
Biblioteca Digital - Facultad de Ciencias Exactas y Naturales (UBA) |
topic |
NIST SRE16 Score Calibration Score Normalization Trial-based Calibration Calibration Speech communication Acoustic conditions Calibration parameters NIST SRE16 Score normalization Speaker clustering Speaker recognition Speaker recognition evaluations Unsupervised techniques Speech recognition |
spellingShingle |
NIST SRE16 Score Calibration Score Normalization Trial-based Calibration Calibration Speech communication Acoustic conditions Calibration parameters NIST SRE16 Score normalization Speaker clustering Speaker recognition Speaker recognition evaluations Unsupervised techniques Speech recognition Castan, D. McLaren, M. Ferrer, L. Lawson, A. Lozano-Diez, A. Lacerda F. Strombergsson S. Wlodarczak M. Heldner M. Gustafson J. House D. Improving robustness of speaker recognition to new conditions using unlabeled data |
topic_facet |
NIST SRE16 Score Calibration Score Normalization Trial-based Calibration Calibration Speech communication Acoustic conditions Calibration parameters NIST SRE16 Score normalization Speaker clustering Speaker recognition Speaker recognition evaluations Unsupervised techniques Speech recognition |
description |
Unsupervised techniques for the adaptation of speaker recognition are important due to the problem of condition mismatch that is prevalent when applying speaker recognition technology to new conditions and the general scarcity of labeled 'in-domain' data. In the recent NIST 2016 Speaker Recognition Evaluation (SRE), symmetric score normalization (Snorm) and calibration using unlabeled in-domain data were shown to be beneficial. Because calibration requires speaker labels for training, speaker-clustering techniques were used to generate pseudo-speakers for learning calibration parameters in those cases where only unlabeled in-domain data was available. These methods performed well in the SRE16. It is unclear, however, whether those techniques generalize well to other data sources. In this work, we benchmark these approaches on several distinctly different databases, after we describe our SRI-CON-UAM team system submission for the NIST 2016 SRE. Our analysis shows that while the benefit of S-norm is also observed across other datasets, applying speaker-clustered calibration provides considerably greater benefit to the system in the context of new acoustic conditions. Copyright © 2017 ISCA. |
format |
CONF |
author |
Castan, D. McLaren, M. Ferrer, L. Lawson, A. Lozano-Diez, A. Lacerda F. Strombergsson S. Wlodarczak M. Heldner M. Gustafson J. House D. |
author_facet |
Castan, D. McLaren, M. Ferrer, L. Lawson, A. Lozano-Diez, A. Lacerda F. Strombergsson S. Wlodarczak M. Heldner M. Gustafson J. House D. |
author_sort |
Castan, D. |
title |
Improving robustness of speaker recognition to new conditions using unlabeled data |
title_short |
Improving robustness of speaker recognition to new conditions using unlabeled data |
title_full |
Improving robustness of speaker recognition to new conditions using unlabeled data |
title_fullStr |
Improving robustness of speaker recognition to new conditions using unlabeled data |
title_full_unstemmed |
Improving robustness of speaker recognition to new conditions using unlabeled data |
title_sort |
improving robustness of speaker recognition to new conditions using unlabeled data |
url |
http://hdl.handle.net/20.500.12110/paper_2308457X_v2017-August_n_p3737_Castan |
work_keys_str_mv |
AT castand improvingrobustnessofspeakerrecognitiontonewconditionsusingunlabeleddata AT mclarenm improvingrobustnessofspeakerrecognitiontonewconditionsusingunlabeleddata AT ferrerl improvingrobustnessofspeakerrecognitiontonewconditionsusingunlabeleddata AT lawsona improvingrobustnessofspeakerrecognitiontonewconditionsusingunlabeleddata AT lozanodieza improvingrobustnessofspeakerrecognitiontonewconditionsusingunlabeleddata AT lacerdaf improvingrobustnessofspeakerrecognitiontonewconditionsusingunlabeleddata AT strombergssons improvingrobustnessofspeakerrecognitiontonewconditionsusingunlabeleddata AT wlodarczakm improvingrobustnessofspeakerrecognitiontonewconditionsusingunlabeleddata AT heldnerm improvingrobustnessofspeakerrecognitiontonewconditionsusingunlabeleddata AT gustafsonj improvingrobustnessofspeakerrecognitiontonewconditionsusingunlabeleddata AT housed improvingrobustnessofspeakerrecognitiontonewconditionsusingunlabeleddata |
_version_ |
1782023941136580608 |