A comparative Re-assessment of feature extractors for deep speaker embeddings

Xuechen Liu; Md Sahidullah; Tomi Kinnunen

doi:10.21437/Interspeech.2020-1765

A comparative Re-assessment of feature extractors for deep speaker embeddings

Xuechen Liu
, Md Sahidullah
, Tomi Kinnunen

University of Eastern Finland
Université de Lorraine

Research output: Chapter in Book or Report/Conference proceeding › Conference Proceeding › peer-review

12 Citations (Scopus)

Abstract

Modern automatic speaker verification relies largely on deep neural networks (DNNs) trained on mel-frequency cepstral coefficient (MFCC) features. While there are alternative feature extraction methods based on phase, prosody and long-term temporal operations, they have not been extensively studied with DNN-based methods. We aim to fill this gap by providing extensive re-assessment of 14 feature extractors on VoxCeleb and SITW datasets. Our findings reveal that features equipped with techniques such as spectral centroids, group delay function, and integrated noise suppression provide promising alternatives to MFCCs for deep speaker embeddings extraction. Experimental results demonstrate up to 16.3% (VoxCeleb) and 25.1% (SITW) relative decrease in equal error rate (EER) to the baseline.

Original language	English
Title of host publication	Interspeech 2020
Publisher	International Speech Communication Association
Pages	3221-3225
Number of pages	5
ISBN (Print)	9781713820697
DOIs	https://doi.org/10.21437/Interspeech.2020-1765
Publication status	Published - 2020
Event	21st Annual Conference of the International Speech Communication Association, INTERSPEECH 2020 - Shanghai, China Duration: 25 Oct 2020 → 29 Oct 2020

Publication series

Name	Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH
Volume	2020-October
ISSN (Print)	2308-457X
ISSN (Electronic)	1990-9772

Conference

Conference	21st Annual Conference of the International Speech Communication Association, INTERSPEECH 2020
Country/Territory	China
City	Shanghai
Period	25/10/20 → 29/10/20

Keywords

Deep speaker embeddings
Feature extraction
Speaker verification

Access to Document

10.21437/Interspeech.2020-1765

Cite this

Liu, X., Sahidullah, M., & Kinnunen, T. (2020). A comparative Re-assessment of feature extractors for deep speaker embeddings. In Interspeech 2020 (pp. 3221-3225). (Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH; Vol. 2020-October). International Speech Communication Association. https://doi.org/10.21437/Interspeech.2020-1765

@inproceedings{85ba5d53135249e386cde261a6e8c8ea,

title = "A comparative Re-assessment of feature extractors for deep speaker embeddings",

abstract = "Modern automatic speaker verification relies largely on deep neural networks (DNNs) trained on mel-frequency cepstral coefficient (MFCC) features. While there are alternative feature extraction methods based on phase, prosody and long-term temporal operations, they have not been extensively studied with DNN-based methods. We aim to fill this gap by providing extensive re-assessment of 14 feature extractors on VoxCeleb and SITW datasets. Our findings reveal that features equipped with techniques such as spectral centroids, group delay function, and integrated noise suppression provide promising alternatives to MFCCs for deep speaker embeddings extraction. Experimental results demonstrate up to 16.3\% (VoxCeleb) and 25.1\% (SITW) relative decrease in equal error rate (EER) to the baseline.",

keywords = "Deep speaker embeddings, Feature extraction, Speaker verification",

author = "Xuechen Liu and Md Sahidullah and Tomi Kinnunen",

note = "Publisher Copyright: {\textcopyright} 2020 International Speech Communication Association. All rights reserved.; 21st Annual Conference of the International Speech Communication Association, INTERSPEECH 2020 ; Conference date: 25-10-2020 Through 29-10-2020",

year = "2020",

doi = "10.21437/Interspeech.2020-1765",

language = "English",

isbn = "9781713820697",

series = "Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH",

publisher = "International Speech Communication Association",

pages = "3221--3225",

booktitle = "Interspeech 2020",

}

Liu, X, Sahidullah, M & Kinnunen, T 2020, A comparative Re-assessment of feature extractors for deep speaker embeddings. in Interspeech 2020. Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH, vol. 2020-October, International Speech Communication Association, pp. 3221-3225, 21st Annual Conference of the International Speech Communication Association, INTERSPEECH 2020, Shanghai, China, 25/10/20. https://doi.org/10.21437/Interspeech.2020-1765

A comparative Re-assessment of feature extractors for deep speaker embeddings. / Liu, Xuechen; Sahidullah, Md; Kinnunen, Tomi.
Interspeech 2020. International Speech Communication Association, 2020. p. 3221-3225 (Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH; Vol. 2020-October).

Research output: Chapter in Book or Report/Conference proceeding › Conference Proceeding › peer-review

TY - GEN

T1 - A comparative Re-assessment of feature extractors for deep speaker embeddings

AU - Liu, Xuechen

AU - Sahidullah, Md

AU - Kinnunen, Tomi

PY - 2020

Y1 - 2020

N2 - Modern automatic speaker verification relies largely on deep neural networks (DNNs) trained on mel-frequency cepstral coefficient (MFCC) features. While there are alternative feature extraction methods based on phase, prosody and long-term temporal operations, they have not been extensively studied with DNN-based methods. We aim to fill this gap by providing extensive re-assessment of 14 feature extractors on VoxCeleb and SITW datasets. Our findings reveal that features equipped with techniques such as spectral centroids, group delay function, and integrated noise suppression provide promising alternatives to MFCCs for deep speaker embeddings extraction. Experimental results demonstrate up to 16.3% (VoxCeleb) and 25.1% (SITW) relative decrease in equal error rate (EER) to the baseline.

AB - Modern automatic speaker verification relies largely on deep neural networks (DNNs) trained on mel-frequency cepstral coefficient (MFCC) features. While there are alternative feature extraction methods based on phase, prosody and long-term temporal operations, they have not been extensively studied with DNN-based methods. We aim to fill this gap by providing extensive re-assessment of 14 feature extractors on VoxCeleb and SITW datasets. Our findings reveal that features equipped with techniques such as spectral centroids, group delay function, and integrated noise suppression provide promising alternatives to MFCCs for deep speaker embeddings extraction. Experimental results demonstrate up to 16.3% (VoxCeleb) and 25.1% (SITW) relative decrease in equal error rate (EER) to the baseline.

KW - Deep speaker embeddings

KW - Feature extraction

KW - Speaker verification

UR - https://www.scopus.com/pages/publications/85098234248

U2 - 10.21437/Interspeech.2020-1765

DO - 10.21437/Interspeech.2020-1765

M3 - Conference Proceeding

AN - SCOPUS:85098234248

SN - 9781713820697

T3 - Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH

SP - 3221

EP - 3225

BT - Interspeech 2020

PB - International Speech Communication Association

T2 - 21st Annual Conference of the International Speech Communication Association, INTERSPEECH 2020

Y2 - 25 October 2020 through 29 October 2020

ER -

A comparative Re-assessment of feature extractors for deep speaker embeddings

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Cite this