Learnable MFCCs for speaker verification

Xuechen Liu; Md Sahidullah; Tomi Kinnunen

doi:10.1109/ISCAS51556.2021.9401593

Learnable MFCCs for speaker verification

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

17 Citations (Scopus)

Abstract

We propose a learnable mel-frequency cepstral coefficients (MFCCs) front-end architecture for deep neural network (DNN) based automatic speaker verification. Our architecture retains the simplicity and interpretability of MFCC-based features while allowing the model to be adapted to data flexibly. In practice, we formulate data-driven version of four linear transforms in a standard MFCC extractor - windowing, discrete Fourier transform (DFT), mel filterbank and discrete cosine transform (DCT). Results reported reach up to 6.7% (VoxCeleb1) and 9.7% (SITW) relative improvement in term of equal error rate (EER) from static MFCCs, without additional tuning effort.

Original language	English
Title of host publication	2021 IEEE International Symposium on Circuits and Systems, ISCAS 2021 - Proceedings
Publisher	Institute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)	9781728192017
DOIs	https://doi.org/10.1109/ISCAS51556.2021.9401593
Publication status	Published - 2021
Event	53rd IEEE International Symposium on Circuits and Systems, ISCAS 2021 - Daegu, Korea, Republic of Duration: 22 May 2021 → 28 May 2021

Publication series

Name	Proceedings - IEEE International Symposium on Circuits and Systems
Volume	2021-May
ISSN (Print)	0271-4310

Conference

Conference	53rd IEEE International Symposium on Circuits and Systems, ISCAS 2021
Country/Territory	Korea, Republic of
City	Daegu
Period	22/05/21 → 28/05/21

Keywords

Feature extraction
Mel-frequency cesptral coefficients (MFCCs)
Speaker verification

Access to Document

10.1109/ISCAS51556.2021.9401593

Cite this

Liu, X., Sahidullah, M., & Kinnunen, T. (2021). Learnable MFCCs for speaker verification. In 2021 IEEE International Symposium on Circuits and Systems, ISCAS 2021 - Proceedings Article 9401593 (Proceedings - IEEE International Symposium on Circuits and Systems; Vol. 2021-May). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/ISCAS51556.2021.9401593

@inproceedings{424c3649583c445a8c04f2306d4139cc,

title = "Learnable MFCCs for speaker verification",

abstract = "We propose a learnable mel-frequency cepstral coefficients (MFCCs) front-end architecture for deep neural network (DNN) based automatic speaker verification. Our architecture retains the simplicity and interpretability of MFCC-based features while allowing the model to be adapted to data flexibly. In practice, we formulate data-driven version of four linear transforms in a standard MFCC extractor - windowing, discrete Fourier transform (DFT), mel filterbank and discrete cosine transform (DCT). Results reported reach up to 6.7\% (VoxCeleb1) and 9.7\% (SITW) relative improvement in term of equal error rate (EER) from static MFCCs, without additional tuning effort.",

keywords = "Feature extraction, Mel-frequency cesptral coefficients (MFCCs), Speaker verification",

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

note = "Publisher Copyright: {\textcopyright} 2021 IEEE; 53rd IEEE International Symposium on Circuits and Systems, ISCAS 2021 ; Conference date: 22-05-2021 Through 28-05-2021",

year = "2021",

doi = "10.1109/ISCAS51556.2021.9401593",

language = "English",

series = "Proceedings - IEEE International Symposium on Circuits and Systems",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

booktitle = "2021 IEEE International Symposium on Circuits and Systems, ISCAS 2021 - Proceedings",

}

Liu, X, Sahidullah, M & Kinnunen, T 2021, Learnable MFCCs for speaker verification. in 2021 IEEE International Symposium on Circuits and Systems, ISCAS 2021 - Proceedings., 9401593, Proceedings - IEEE International Symposium on Circuits and Systems, vol. 2021-May, Institute of Electrical and Electronics Engineers Inc., 53rd IEEE International Symposium on Circuits and Systems, ISCAS 2021, Daegu, Korea, Republic of, 22/05/21. https://doi.org/10.1109/ISCAS51556.2021.9401593

Learnable MFCCs for speaker verification. / Liu, Xuechen; Sahidullah, Md; Kinnunen, Tomi.
2021 IEEE International Symposium on Circuits and Systems, ISCAS 2021 - Proceedings. Institute of Electrical and Electronics Engineers Inc., 2021. 9401593 (Proceedings - IEEE International Symposium on Circuits and Systems; Vol. 2021-May).

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

TY - GEN

T1 - Learnable MFCCs for speaker verification

AU - Liu, Xuechen

AU - Sahidullah, Md

AU - Kinnunen, Tomi

PY - 2021

Y1 - 2021

N2 - We propose a learnable mel-frequency cepstral coefficients (MFCCs) front-end architecture for deep neural network (DNN) based automatic speaker verification. Our architecture retains the simplicity and interpretability of MFCC-based features while allowing the model to be adapted to data flexibly. In practice, we formulate data-driven version of four linear transforms in a standard MFCC extractor - windowing, discrete Fourier transform (DFT), mel filterbank and discrete cosine transform (DCT). Results reported reach up to 6.7% (VoxCeleb1) and 9.7% (SITW) relative improvement in term of equal error rate (EER) from static MFCCs, without additional tuning effort.

AB - We propose a learnable mel-frequency cepstral coefficients (MFCCs) front-end architecture for deep neural network (DNN) based automatic speaker verification. Our architecture retains the simplicity and interpretability of MFCC-based features while allowing the model to be adapted to data flexibly. In practice, we formulate data-driven version of four linear transforms in a standard MFCC extractor - windowing, discrete Fourier transform (DFT), mel filterbank and discrete cosine transform (DCT). Results reported reach up to 6.7% (VoxCeleb1) and 9.7% (SITW) relative improvement in term of equal error rate (EER) from static MFCCs, without additional tuning effort.

KW - Feature extraction

KW - Mel-frequency cesptral coefficients (MFCCs)

KW - Speaker verification

UR - https://www.scopus.com/pages/publications/85109019883

U2 - 10.1109/ISCAS51556.2021.9401593

DO - 10.1109/ISCAS51556.2021.9401593

M3 - Conference Proceeding

AN - SCOPUS:85109019883

T3 - Proceedings - IEEE International Symposium on Circuits and Systems

BT - 2021 IEEE International Symposium on Circuits and Systems, ISCAS 2021 - Proceedings

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 53rd IEEE International Symposium on Circuits and Systems, ISCAS 2021

Y2 - 22 May 2021 through 28 May 2021

ER -

Learnable MFCCs for speaker verification

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Cite this