Speaker Verification Experiments for Adults and Children Using Shared Embedding Spaces

Tuomas Kaseva; Hemant Kathania; Aku Rouhe; Mikko Kurimo

Speaker Verification Experiments for Adults and Children Using Shared Embedding Spaces

Tuomas Kaseva, Hemant Kathania, Aku Rouhe, Mikko Kurimo

Speech Recognition

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Scientific › peer-review

18 Downloads (Pure)

Abstract

For children, the system trained on a large corpus of adult speakers performed worse than a system trained on a much smaller corpus of children’s speech. This is due to the acoustic mismatch between training and testing data. To capture more acoustic variability we trained a shared system with mixed data from adults and children. The shared system yields the best EER for children with no degradation for adults. Thus, the single system trained with mixed data is applicable for speaker verification for both adults and children.

Original language	English
Title of host publication	Proceedings of the 23rd Nordic Conference on Computational Linguistics (NoDaLiDa), May 31-June 2, 2021
Place of Publication	Sweden
Publisher	Linköping University Electronic Press
Pages	86-93
Number of pages	8
ISBN (Electronic)	978-91-7929-614-8
Publication status	Published - 2021
MoE publication type	A4 Conference publication
Event	Nordic Conference on Computational Linguistics - Reykjavik, Iceland Duration: 31 May 2021 → 2 Jun 2021

Publication series

Name	Linköping Electronic Conference Proceedings
Number	178
ISSN (Print)	1650-3686
ISSN (Electronic)	1650-3740

Conference

Conference	Nordic Conference on Computational Linguistics
Abbreviated title	NoDaLiDa
Country/Territory	Iceland
City	Reykjavik
Period	31/05/2021 → 02/06/2021

Access to Document

Kaseva_Speaker_verification_experiments_adults_children_Nodalida2021Final published version, 272 KB

https://ep.liu.se/en/conference-article.aspx?series=ecp&issue=178&Article_No=9

OpenUrl availability

Full text

MeMAD: Methods for Managing Audiovisual Data: Combining Automatic Efficiency with Human Accuracy
Kurimo, M., Grönroos, S., Brander, T., Porjazovski, D., Raitio, R., Rouhe, A., Grósz, T. & Virkkunen, A.
27/12/2017 → 31/03/2021
Project: EU: Framework programmes funding

Cite this

Kaseva, T., Kathania, H., Rouhe, A., & Kurimo, M. (2021). Speaker Verification Experiments for Adults and Children Using Shared Embedding Spaces. In Proceedings of the 23rd Nordic Conference on Computational Linguistics (NoDaLiDa), May 31-June 2, 2021 (pp. 86-93). (Linköping Electronic Conference Proceedings; No. 178). Linköping University Electronic Press. https://ep.liu.se/en/conference-article.aspx?series=ecp&issue=178&Article_No=9

@inproceedings{25d15f7d4a034174bc9a236421c46329,

title = "Speaker Verification Experiments for Adults and Children Using Shared Embedding Spaces",

abstract = "For children, the system trained on a large corpus of adult speakers performed worse than a system trained on a much smaller corpus of children{\textquoteright}s speech. This is due to the acoustic mismatch between training and testing data. To capture more acoustic variability we trained a shared system with mixed data from adults and children. The shared system yields the best EER for children with no degradation for adults. Thus, the single system trained with mixed data is applicable for speaker verification for both adults and children.",

author = "Tuomas Kaseva and Hemant Kathania and Aku Rouhe and Mikko Kurimo",

note = "| openaire: EC/H2020/780069/EU//MeMAD; Nordic Conference on Computational Linguistics, NoDaLiDa ; Conference date: 31-05-2021 Through 02-06-2021",

year = "2021",

language = "English",

series = "Link{\"o}ping Electronic Conference Proceedings",

publisher = "Link{\"o}ping University Electronic Press",

number = "178",

pages = "86--93",

booktitle = "Proceedings of the 23rd Nordic Conference on Computational Linguistics (NoDaLiDa), May 31-June 2, 2021",

address = "Sweden",

}

Kaseva, T, Kathania, H, Rouhe, A & Kurimo, M 2021, Speaker Verification Experiments for Adults and Children Using Shared Embedding Spaces. in Proceedings of the 23rd Nordic Conference on Computational Linguistics (NoDaLiDa), May 31-June 2, 2021. Linköping Electronic Conference Proceedings, no. 178, Linköping University Electronic Press, Sweden, pp. 86-93, Nordic Conference on Computational Linguistics, Reykjavik, Iceland, 31/05/2021. <https://ep.liu.se/en/conference-article.aspx?series=ecp&issue=178&Article_No=9>

Speaker Verification Experiments for Adults and Children Using Shared Embedding Spaces. / Kaseva, Tuomas; Kathania, Hemant; Rouhe, Aku et al.
Proceedings of the 23rd Nordic Conference on Computational Linguistics (NoDaLiDa), May 31-June 2, 2021. Sweden: Linköping University Electronic Press, 2021. p. 86-93 (Linköping Electronic Conference Proceedings; No. 178).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Scientific › peer-review

TY - GEN

T1 - Speaker Verification Experiments for Adults and Children Using Shared Embedding Spaces

AU - Kaseva, Tuomas

AU - Kathania, Hemant

AU - Rouhe, Aku

AU - Kurimo, Mikko

N1 - | openaire: EC/H2020/780069/EU//MeMAD

PY - 2021

Y1 - 2021

N2 - For children, the system trained on a large corpus of adult speakers performed worse than a system trained on a much smaller corpus of children’s speech. This is due to the acoustic mismatch between training and testing data. To capture more acoustic variability we trained a shared system with mixed data from adults and children. The shared system yields the best EER for children with no degradation for adults. Thus, the single system trained with mixed data is applicable for speaker verification for both adults and children.

AB - For children, the system trained on a large corpus of adult speakers performed worse than a system trained on a much smaller corpus of children’s speech. This is due to the acoustic mismatch between training and testing data. To capture more acoustic variability we trained a shared system with mixed data from adults and children. The shared system yields the best EER for children with no degradation for adults. Thus, the single system trained with mixed data is applicable for speaker verification for both adults and children.

M3 - Conference contribution

T3 - Linköping Electronic Conference Proceedings

SP - 86

EP - 93

BT - Proceedings of the 23rd Nordic Conference on Computational Linguistics (NoDaLiDa), May 31-June 2, 2021

PB - Linköping University Electronic Press

CY - Sweden

T2 - Nordic Conference on Computational Linguistics

Y2 - 31 May 2021 through 2 June 2021

ER -

Speaker Verification Experiments for Adults and Children Using Shared Embedding Spaces

Abstract

Publication series

Conference

Access to Document

OpenUrl availability

Fingerprint

Projects

MeMAD: Methods for Managing Audiovisual Data: Combining Automatic Efficiency with Human Accuracy

Cite this