Speech Enhancement by Classification of Noisy Signals Decomposed Using NMF and Wiener Filtering

Research output: Contribution to book/anthology/report/conference proceedingArticle in proceedingResearchpeer-review

3 Citations (Scopus)
124 Downloads (Pure)

Abstract

Supervised non-negative matrix factorization (NMF) is effective in speech enhancement through training spectral models of speech and noise signals. However, the enhancement quality reduces when the models are trained on data that is not highly relevant to a speech signal and a noise signal in a noisy observation. In this paper, we propose to train a classifier in order to overcome such poor characterization of the signals through the trained models. The main idea is to decompose the noisy observation into parts and the enhanced signal is reconstructed by combining the less-corrupted ones which are identified in the cepstral domain using the trained classifier. We apply unsupervised NMF followed by Wiener filtering for the decomposition, and use a support vector machine trained on the mel-frequency cepstral coefficients of the parts of training speech and noise signals for the classification. The results show the effectiveness of the proposed method compared with the supervised NMF.
Original languageEnglish
Title of host publication26th European Signal Processing Conference (EUSIPCO)
Number of pages5
PublisherIEEE
Publication date2018
Article number8553123
ISBN (Electronic)978-9-0827-9701-5
DOIs
Publication statusPublished - 2018
Event26th European Signal Processing Conference (EUSIPCO 2018) - Rome, Italy
Duration: 3 Sep 20187 Sep 2018
Conference number: 26
http://www.eusipco2018.org

Conference

Conference26th European Signal Processing Conference (EUSIPCO 2018)
Number26
CountryItaly
CityRome
Period03/09/201807/09/2018
Internet address
SeriesProceedings of the European Signal Processing Conference
ISSN2076-1465

Fingerprint

Speech enhancement
Factorization
Classifiers
Support vector machines
Decomposition

Cite this

Fakhry, M., Poorjam, A. H., & Christensen, M. G. (2018). Speech Enhancement by Classification of Noisy Signals Decomposed Using NMF and Wiener Filtering. In 26th European Signal Processing Conference (EUSIPCO) [8553123] IEEE. Proceedings of the European Signal Processing Conference https://doi.org/10.23919/EUSIPCO.2018.8553123
Fakhry, Mahmoud ; Poorjam, Amir Hossein ; Christensen, Mads Græsbøll. / Speech Enhancement by Classification of Noisy Signals Decomposed Using NMF and Wiener Filtering. 26th European Signal Processing Conference (EUSIPCO). IEEE, 2018. (Proceedings of the European Signal Processing Conference).
@inproceedings{44738e4d9fec4a38a508c1fb8031a7d0,
title = "Speech Enhancement by Classification of Noisy Signals Decomposed Using NMF and Wiener Filtering",
abstract = "Supervised non-negative matrix factorization (NMF) is effective in speech enhancement through training spectral models of speech and noise signals. However, the enhancement quality reduces when the models are trained on data that is not highly relevant to a speech signal and a noise signal in a noisy observation. In this paper, we propose to train a classifier in order to overcome such poor characterization of the signals through the trained models. The main idea is to decompose the noisy observation into parts and the enhanced signal is reconstructed by combining the less-corrupted ones which are identified in the cepstral domain using the trained classifier. We apply unsupervised NMF followed by Wiener filtering for the decomposition, and use a support vector machine trained on the mel-frequency cepstral coefficients of the parts of training speech and noise signals for the classification. The results show the effectiveness of the proposed method compared with the supervised NMF.",
author = "Mahmoud Fakhry and Poorjam, {Amir Hossein} and Christensen, {Mads Gr{\ae}sb{\o}ll}",
year = "2018",
doi = "10.23919/EUSIPCO.2018.8553123",
language = "English",
series = "Proceedings of the European Signal Processing Conference",
publisher = "IEEE",
booktitle = "26th European Signal Processing Conference (EUSIPCO)",
address = "United States",

}

Fakhry, M, Poorjam, AH & Christensen, MG 2018, Speech Enhancement by Classification of Noisy Signals Decomposed Using NMF and Wiener Filtering. in 26th European Signal Processing Conference (EUSIPCO)., 8553123, IEEE, Proceedings of the European Signal Processing Conference, 26th European Signal Processing Conference (EUSIPCO 2018), Rome, Italy, 03/09/2018. https://doi.org/10.23919/EUSIPCO.2018.8553123

Speech Enhancement by Classification of Noisy Signals Decomposed Using NMF and Wiener Filtering. / Fakhry, Mahmoud; Poorjam, Amir Hossein; Christensen, Mads Græsbøll.

26th European Signal Processing Conference (EUSIPCO). IEEE, 2018. 8553123 (Proceedings of the European Signal Processing Conference).

Research output: Contribution to book/anthology/report/conference proceedingArticle in proceedingResearchpeer-review

TY - GEN

T1 - Speech Enhancement by Classification of Noisy Signals Decomposed Using NMF and Wiener Filtering

AU - Fakhry, Mahmoud

AU - Poorjam, Amir Hossein

AU - Christensen, Mads Græsbøll

PY - 2018

Y1 - 2018

N2 - Supervised non-negative matrix factorization (NMF) is effective in speech enhancement through training spectral models of speech and noise signals. However, the enhancement quality reduces when the models are trained on data that is not highly relevant to a speech signal and a noise signal in a noisy observation. In this paper, we propose to train a classifier in order to overcome such poor characterization of the signals through the trained models. The main idea is to decompose the noisy observation into parts and the enhanced signal is reconstructed by combining the less-corrupted ones which are identified in the cepstral domain using the trained classifier. We apply unsupervised NMF followed by Wiener filtering for the decomposition, and use a support vector machine trained on the mel-frequency cepstral coefficients of the parts of training speech and noise signals for the classification. The results show the effectiveness of the proposed method compared with the supervised NMF.

AB - Supervised non-negative matrix factorization (NMF) is effective in speech enhancement through training spectral models of speech and noise signals. However, the enhancement quality reduces when the models are trained on data that is not highly relevant to a speech signal and a noise signal in a noisy observation. In this paper, we propose to train a classifier in order to overcome such poor characterization of the signals through the trained models. The main idea is to decompose the noisy observation into parts and the enhanced signal is reconstructed by combining the less-corrupted ones which are identified in the cepstral domain using the trained classifier. We apply unsupervised NMF followed by Wiener filtering for the decomposition, and use a support vector machine trained on the mel-frequency cepstral coefficients of the parts of training speech and noise signals for the classification. The results show the effectiveness of the proposed method compared with the supervised NMF.

U2 - 10.23919/EUSIPCO.2018.8553123

DO - 10.23919/EUSIPCO.2018.8553123

M3 - Article in proceeding

T3 - Proceedings of the European Signal Processing Conference

BT - 26th European Signal Processing Conference (EUSIPCO)

PB - IEEE

ER -

Fakhry M, Poorjam AH, Christensen MG. Speech Enhancement by Classification of Noisy Signals Decomposed Using NMF and Wiener Filtering. In 26th European Signal Processing Conference (EUSIPCO). IEEE. 2018. 8553123. (Proceedings of the European Signal Processing Conference). https://doi.org/10.23919/EUSIPCO.2018.8553123