A new phase-based feature representation for robust speech recognition

E. Loweimi; S. M. Ahadi; T. Drugman

doi:10.1109/ICASSP.2013.6639051

A new phase-based feature representation for robust speech recognition

E. Loweimi, S. M. Ahadi, T. Drugman

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract

The aim of this paper is to introduce a novel phase-based feature representation for robust speech recognition. This method consists of four main parts: autoregressive (AR) model extraction, group delay function (GDF) computation, compression, and scale information augmentation. Coupling GDF with an AR model results in a high-resolution estimate of the power spectrum with low frequency leakage. The compression step includes two stages similar to MFCC without taking a logarithm of the output energies. The fourth part augments the phase-based feature vector with scale information which is based on the Hilbert transform relations and complements the phase spectrum information. In the presence of additive and convolutional noises, the proposed method has led to 15% and 12% reductions in the averaged error rates, respectively (SNR ranging from 0 to 20 dB), compared to the standard MFCCs.

Original language	English
Title of host publication	2013 IEEE International Conference on Acoustics, Speech and Signal Processing
Publisher	Institute of Electrical and Electronics Engineers
Pages	7155-7159
Number of pages	5
ISBN (Electronic)	978-1-4799-0356-6
DOIs	https://doi.org/10.1109/ICASSP.2013.6639051
Publication status	Published - 1 May 2013
Event	38th IEEE International Conference on Acoustics, Speech, and Signal Processing - Vancouver, Canada Duration: 26 May 2013 → 31 May 2013 https://www2.securecms.com/ICASSP2013/default.asp

Conference

Conference	38th IEEE International Conference on Acoustics, Speech, and Signal Processing
Abbreviated title	ICASSP 2013
Country/Territory	Canada
City	Vancouver
Period	26/05/13 → 31/05/13
Internet address	https://www2.securecms.com/ICASSP2013/default.asp

Keywords / Materials (for Non-textual outputs)

autoregressive processes
error statistics
feature extraction
Hilbert transforms
image representation
speech recognition
phase-based feature representation
robust speech recognition
AR model extraction
autoregressive model extraction
group delay function computation
GDF computation
scale information augmentation
power spectrum
high-resolution estimation
low frequency leakage
compression step
feature vector
Hilbert transform relations
phase spectrum information
additive noises
convolutional noises
averaged error rates
standard MFCC
Speech
Robustness
Speech recognition
Abstracts
Mel frequency cepstral coefficient
Speech phase spectrum
group delay
compression
scale information

Access to Document

10.1109/ICASSP.2013.6639051

https://ieeexplore.ieee.org/document/6639051

Cite this

@inproceedings{bcee492e9c4e4f8e84c30601df6ff1b5,

title = "A new phase-based feature representation for robust speech recognition",

abstract = "The aim of this paper is to introduce a novel phase-based feature representation for robust speech recognition. This method consists of four main parts: autoregressive (AR) model extraction, group delay function (GDF) computation, compression, and scale information augmentation. Coupling GDF with an AR model results in a high-resolution estimate of the power spectrum with low frequency leakage. The compression step includes two stages similar to MFCC without taking a logarithm of the output energies. The fourth part augments the phase-based feature vector with scale information which is based on the Hilbert transform relations and complements the phase spectrum information. In the presence of additive and convolutional noises, the proposed method has led to 15% and 12% reductions in the averaged error rates, respectively (SNR ranging from 0 to 20 dB), compared to the standard MFCCs.",

keywords = "autoregressive processes, error statistics, feature extraction, Hilbert transforms, image representation, speech recognition, phase-based feature representation, robust speech recognition, AR model extraction, autoregressive model extraction, group delay function computation, GDF computation, scale information augmentation, power spectrum, high-resolution estimation, low frequency leakage, compression step, feature vector, Hilbert transform relations, phase spectrum information, additive noises, convolutional noises, averaged error rates, standard MFCC, Speech, Robustness, Speech recognition, Abstracts, Mel frequency cepstral coefficient, Speech phase spectrum, group delay, compression, scale information",

author = "E. Loweimi and Ahadi, {S. M.} and T. Drugman",

year = "2013",

month = may,

day = "1",

doi = "10.1109/ICASSP.2013.6639051",

language = "English",

pages = "7155--7159",

booktitle = "2013 IEEE International Conference on Acoustics, Speech and Signal Processing",

publisher = "Institute of Electrical and Electronics Engineers",

address = "United States",

note = "38th IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2013 ; Conference date: 26-05-2013 Through 31-05-2013",

url = "https://www2.securecms.com/ICASSP2013/default.asp",

}

Loweimi, E, Ahadi, SM & Drugman, T 2013, A new phase-based feature representation for robust speech recognition. in 2013 IEEE International Conference on Acoustics, Speech and Signal Processing. Institute of Electrical and Electronics Engineers, pp. 7155-7159, 38th IEEE International Conference on Acoustics, Speech, and Signal Processing, Vancouver, British Columbia, Canada, 26/05/13. https://doi.org/10.1109/ICASSP.2013.6639051

TY - GEN

T1 - A new phase-based feature representation for robust speech recognition

AU - Loweimi, E.

AU - Ahadi, S. M.

AU - Drugman, T.

PY - 2013/5/1

Y1 - 2013/5/1

N2 - The aim of this paper is to introduce a novel phase-based feature representation for robust speech recognition. This method consists of four main parts: autoregressive (AR) model extraction, group delay function (GDF) computation, compression, and scale information augmentation. Coupling GDF with an AR model results in a high-resolution estimate of the power spectrum with low frequency leakage. The compression step includes two stages similar to MFCC without taking a logarithm of the output energies. The fourth part augments the phase-based feature vector with scale information which is based on the Hilbert transform relations and complements the phase spectrum information. In the presence of additive and convolutional noises, the proposed method has led to 15% and 12% reductions in the averaged error rates, respectively (SNR ranging from 0 to 20 dB), compared to the standard MFCCs.

AB - The aim of this paper is to introduce a novel phase-based feature representation for robust speech recognition. This method consists of four main parts: autoregressive (AR) model extraction, group delay function (GDF) computation, compression, and scale information augmentation. Coupling GDF with an AR model results in a high-resolution estimate of the power spectrum with low frequency leakage. The compression step includes two stages similar to MFCC without taking a logarithm of the output energies. The fourth part augments the phase-based feature vector with scale information which is based on the Hilbert transform relations and complements the phase spectrum information. In the presence of additive and convolutional noises, the proposed method has led to 15% and 12% reductions in the averaged error rates, respectively (SNR ranging from 0 to 20 dB), compared to the standard MFCCs.

KW - autoregressive processes

KW - error statistics

KW - feature extraction

KW - Hilbert transforms

KW - image representation

KW - speech recognition

KW - phase-based feature representation

KW - robust speech recognition

KW - AR model extraction

KW - autoregressive model extraction

KW - group delay function computation

KW - GDF computation

KW - scale information augmentation

KW - power spectrum

KW - high-resolution estimation

KW - low frequency leakage

KW - compression step

KW - feature vector

KW - Hilbert transform relations

KW - phase spectrum information

KW - additive noises

KW - convolutional noises

KW - averaged error rates

KW - standard MFCC

KW - Speech

KW - Robustness

KW - Speech recognition

KW - Abstracts

KW - Mel frequency cepstral coefficient

KW - Speech phase spectrum

KW - group delay

KW - compression

KW - scale information

U2 - 10.1109/ICASSP.2013.6639051

DO - 10.1109/ICASSP.2013.6639051

M3 - Conference contribution

SP - 7155

EP - 7159

BT - 2013 IEEE International Conference on Acoustics, Speech and Signal Processing

PB - Institute of Electrical and Electronics Engineers

T2 - 38th IEEE International Conference on Acoustics, Speech, and Signal Processing

Y2 - 26 May 2013 through 31 May 2013

ER -

A new phase-based feature representation for robust speech recognition

Abstract

Conference

Keywords / Materials (for Non-textual outputs)

Access to Document

Fingerprint

Cite this