Skip to main content

Lund University Publications

LUND UNIVERSITY LIBRARIES

Low-Variance Multitaper MFCC Features: A Case Study in Robust Speaker Verification

Kinnunen, Tomi ; Saeidi, Rahim ; Sedlak, Filip ; Lee, Kong Aik ; Sandberg, Johan LU ; Sandsten, Maria LU and Li, Haizhou (2012) In IEEE Transactions on Audio, Speech, and Language Processing 20(7). p.1990-2001
Abstract
In speech and audio applications, short-term signal spectrum is often represented using mel-frequency cepstral coefficients (MFCCs) computed from a windowed discrete Fourier transform (DFT). Windowing reduces spectral leakage but variance of the spectrum estimate remains high. An elegant extension to windowed DFT is the so-called multitaper method which uses multiple time-domain windows (tapers) with frequency-domain averaging. Multitapers have received little attention in speech processing even though they produce low-variance features. In this paper, we propose the multitaper method for MFCC extraction with a practical focus. We provide, first, detailed statistical analysis of MFCC bias and variance using autoregressive process... (More)
In speech and audio applications, short-term signal spectrum is often represented using mel-frequency cepstral coefficients (MFCCs) computed from a windowed discrete Fourier transform (DFT). Windowing reduces spectral leakage but variance of the spectrum estimate remains high. An elegant extension to windowed DFT is the so-called multitaper method which uses multiple time-domain windows (tapers) with frequency-domain averaging. Multitapers have received little attention in speech processing even though they produce low-variance features. In this paper, we propose the multitaper method for MFCC extraction with a practical focus. We provide, first, detailed statistical analysis of MFCC bias and variance using autoregressive process simulations on the TIMIT corpus. For speaker verification experiments on the NIST 2002 and 2008 SRE corpora, we consider three Gaussian mixture model based classifiers with universal background model (GMM-UBM), support vector machine (GMM-SVM) and joint factor analysis (GMM-JFA). Multitapers improve MinDCF over the baseline windowed DFT by relative 20.4% (GMM-SVM) and 13.7% (GMM-JFA) on the interview-interview condition in NIST 2008. The GMM-JFA system further reduces MinDCF by 18.7% on the telephone data. With these improvements and generally noncritical parameter selection, multitaper MFCCs are a viable candidate for replacing the conventional MFCCs. (Less)
Please use this url to cite or link to this publication:
author
; ; ; ; ; and
organization
publishing date
type
Contribution to journal
publication status
published
subject
keywords
Mel-frequency cepstral coefficient (MFCC), multitaper, small-variance, estimation, speaker verification
in
IEEE Transactions on Audio, Speech, and Language Processing
volume
20
issue
7
pages
1990 - 2001
publisher
IEEE - Institute of Electrical and Electronics Engineers Inc.
external identifiers
  • wos:000303893000003
  • scopus:84860850285
ISSN
1558-7924
DOI
10.1109/TASL.2012.2191960
language
English
LU publication?
yes
id
e055865e-5dbc-4a3f-8a32-1629664eca7f (old id 2826390)
date added to LUP
2016-04-01 10:00:55
date last changed
2022-04-12 01:09:22
@article{e055865e-5dbc-4a3f-8a32-1629664eca7f,
  abstract     = {{In speech and audio applications, short-term signal spectrum is often represented using mel-frequency cepstral coefficients (MFCCs) computed from a windowed discrete Fourier transform (DFT). Windowing reduces spectral leakage but variance of the spectrum estimate remains high. An elegant extension to windowed DFT is the so-called multitaper method which uses multiple time-domain windows (tapers) with frequency-domain averaging. Multitapers have received little attention in speech processing even though they produce low-variance features. In this paper, we propose the multitaper method for MFCC extraction with a practical focus. We provide, first, detailed statistical analysis of MFCC bias and variance using autoregressive process simulations on the TIMIT corpus. For speaker verification experiments on the NIST 2002 and 2008 SRE corpora, we consider three Gaussian mixture model based classifiers with universal background model (GMM-UBM), support vector machine (GMM-SVM) and joint factor analysis (GMM-JFA). Multitapers improve MinDCF over the baseline windowed DFT by relative 20.4% (GMM-SVM) and 13.7% (GMM-JFA) on the interview-interview condition in NIST 2008. The GMM-JFA system further reduces MinDCF by 18.7% on the telephone data. With these improvements and generally noncritical parameter selection, multitaper MFCCs are a viable candidate for replacing the conventional MFCCs.}},
  author       = {{Kinnunen, Tomi and Saeidi, Rahim and Sedlak, Filip and Lee, Kong Aik and Sandberg, Johan and Sandsten, Maria and Li, Haizhou}},
  issn         = {{1558-7924}},
  keywords     = {{Mel-frequency cepstral coefficient (MFCC); multitaper; small-variance; estimation; speaker verification}},
  language     = {{eng}},
  number       = {{7}},
  pages        = {{1990--2001}},
  publisher    = {{IEEE - Institute of Electrical and Electronics Engineers Inc.}},
  series       = {{IEEE Transactions on Audio, Speech, and Language Processing}},
  title        = {{Low-Variance Multitaper MFCC Features: A Case Study in Robust Speaker Verification}},
  url          = {{http://dx.doi.org/10.1109/TASL.2012.2191960}},
  doi          = {{10.1109/TASL.2012.2191960}},
  volume       = {{20}},
  year         = {{2012}},
}