Discriminative Enhancement for Single Channel Audio Source Separation Using Deep Neural Networks

Emad M Grais; Gerard Roma; Andrew JR Simpson; Mark D Plumbley

doi:10.1007/978-3-319-53547-0_23

Discriminative Enhancement for Single Channel Audio Source Separation Using Deep Neural Networks

Emad M Grais, Gerard Roma, Andrew JR Simpson, Mark D Plumbley

University of Surrey

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

8 Citations (Scopus)

Abstract

The sources separated by most single channel audio source separation techniques are usually distorted and each separated source contains residual signals from the other sources. To tackle this problem, we propose to enhance the separated sources to decrease the distortion and interference between the separated sources using deep neural networks (DNNs). Two different DNNs are used in this work. The first DNN is used to separate the sources from the mixed signal. The second DNN is used to enhance the separated signals. To consider the interactions between the separated sources, we propose to use a single DNN to enhance all the separated sources together. To reduce the residual signals of one source from the other separated sources (interference), we train the DNN for enhancement discriminatively to maximize the dissimilarity between the predicted sources. The experimental results show that using discriminative enhancement decreases the distortion and interference between the separated sources.

Original language	English
Title of host publication	Latent Variable Analysis and Signal Separation
Editors	Petr Tichavský, Massoud Babaie-Zadeh, Olivier J.J. Michel, Nadège Thirion-Moreau
Publisher	Springer, Cham
Pages	236-246
Number of pages	11
ISBN (Electronic)	9783319535470
ISBN (Print)	9783319535463
DOIs	https://doi.org/10.1007/978-3-319-53547-0_23
Publication status	Published - 15 Feb 2017
Externally published	Yes
Event	13th International Conference on Latent Variable Analysis and Signal Separation - Grenoble-Alpes University, Grenoble, France Duration: 21 Feb 2017 → 23 Feb 2017 Conference number: 13 http://www.lva-ica-2017.com/ (Link to Conference Website)

Publication series

Name	Lecture Notes in Computer Science
Publisher	Springer
Volume	10169
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	13th International Conference on Latent Variable Analysis and Signal Separation
Abbreviated title	LVA/ICA 2017
Country/Territory	France
City	Grenoble
Period	21/02/17 → 23/02/17
Internet address	http://www.lva-ica-2017.com/ (Link to Conference Website)

Access to Document

10.1007/978-3-319-53547-0_23Licence: Unspecified

http://epubs.surrey.ac.uk/813112/Licence: Unspecified

Improving single-network single-channel separation of musical audio with convolutional layers
Roma, G., Green, O. & Tremblay, P. A., 6 Jun 2018, Latent Variable Analysis and Signal Separation: 14th International Conference, LVA/ICA 2018, Guildford, UK, July 2–5, 2018, Proceedings. Gannot, S., Deville, Y., Mason, R., Plumbley, M. D. & Ward, D. (eds.). Springer Verlag, p. 306-315 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); vol. 10891 LNCS).
Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Open Access
File
6 Link opens in a new tab Citations (Scopus)
Combining mask estimates for single channel audio source separation using deep neural networks
Grais, E. M., Roma, G., Simpson, A. J. & Plumbley, M. D., Sept 2016, Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH. p. 3339-3343 5 p.
Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Open Access
23 Link opens in a new tab Citations (Scopus)

Cite this

Grais, E. M., Roma, G., Simpson, A. JR., & Plumbley, M. D. (2017). Discriminative Enhancement for Single Channel Audio Source Separation Using Deep Neural Networks. In P. Tichavský, M. Babaie-Zadeh, O. J. J. Michel, & N. Thirion-Moreau (Eds.), Latent Variable Analysis and Signal Separation (pp. 236-246). (Lecture Notes in Computer Science; Vol. 10169). Springer, Cham. https://doi.org/10.1007/978-3-319-53547-0_23

Grais, Emad M ; Roma, Gerard ; Simpson, Andrew JR et al. / Discriminative Enhancement for Single Channel Audio Source Separation Using Deep Neural Networks. Latent Variable Analysis and Signal Separation. editor / Petr Tichavský ; Massoud Babaie-Zadeh ; Olivier J.J. Michel ; Nadège Thirion-Moreau. Springer, Cham, 2017. pp. 236-246 (Lecture Notes in Computer Science).

@inproceedings{1e62f770e5fd4148a87f11a2b2ef3d24,

title = "Discriminative Enhancement for Single Channel Audio Source Separation Using Deep Neural Networks",

abstract = "The sources separated by most single channel audio source separation techniques are usually distorted and each separated source contains residual signals from the other sources. To tackle this problem, we propose to enhance the separated sources to decrease the distortion and interference between the separated sources using deep neural networks (DNNs). Two different DNNs are used in this work. The first DNN is used to separate the sources from the mixed signal. The second DNN is used to enhance the separated signals. To consider the interactions between the separated sources, we propose to use a single DNN to enhance all the separated sources together. To reduce the residual signals of one source from the other separated sources (interference), we train the DNN for enhancement discriminatively to maximize the dissimilarity between the predicted sources. The experimental results show that using discriminative enhancement decreases the distortion and interference between the separated sources.",

keywords = "Audio enhancement, Deep neural networks, Discriminative training, Single channel audio source separation",

author = "Grais, \{Emad M\} and Gerard Roma and Simpson, \{Andrew JR\} and Plumbley, \{Mark D\}",

year = "2017",

month = feb,

day = "15",

doi = "10.1007/978-3-319-53547-0\_23",

language = "English",

isbn = "9783319535463",

series = "Lecture Notes in Computer Science",

publisher = "Springer, Cham",

pages = "236--246",

editor = "Petr Tichavsk{\'y} and Massoud Babaie-Zadeh and Michel, \{Olivier J.J.\} and Nad{\`e}ge Thirion-Moreau",

booktitle = "Latent Variable Analysis and Signal Separation",

address = "Switzerland",

note = "13th International Conference on Latent Variable Analysis and Signal Separation, LVA/ICA 2017 ; Conference date: 21-02-2017 Through 23-02-2017",

url = "http://www.lva-ica-2017.com/",

}

Grais, EM, Roma, G, Simpson, AJR & Plumbley, MD 2017, Discriminative Enhancement for Single Channel Audio Source Separation Using Deep Neural Networks. in P Tichavský, M Babaie-Zadeh, OJJ Michel & N Thirion-Moreau (eds), Latent Variable Analysis and Signal Separation. Lecture Notes in Computer Science, vol. 10169, Springer, Cham, pp. 236-246, 13th International Conference on Latent Variable Analysis and Signal Separation, Grenoble, France, 21/02/17. https://doi.org/10.1007/978-3-319-53547-0_23

Discriminative Enhancement for Single Channel Audio Source Separation Using Deep Neural Networks. / Grais, Emad M; Roma, Gerard; Simpson, Andrew JR et al.
Latent Variable Analysis and Signal Separation. ed. / Petr Tichavský; Massoud Babaie-Zadeh; Olivier J.J. Michel; Nadège Thirion-Moreau. Springer, Cham, 2017. p. 236-246 (Lecture Notes in Computer Science; Vol. 10169).

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

TY - GEN

T1 - Discriminative Enhancement for Single Channel Audio Source Separation Using Deep Neural Networks

AU - Grais, Emad M

AU - Roma, Gerard

AU - Simpson, Andrew JR

AU - Plumbley, Mark D

N1 - Conference code: 13

PY - 2017/2/15

Y1 - 2017/2/15

N2 - The sources separated by most single channel audio source separation techniques are usually distorted and each separated source contains residual signals from the other sources. To tackle this problem, we propose to enhance the separated sources to decrease the distortion and interference between the separated sources using deep neural networks (DNNs). Two different DNNs are used in this work. The first DNN is used to separate the sources from the mixed signal. The second DNN is used to enhance the separated signals. To consider the interactions between the separated sources, we propose to use a single DNN to enhance all the separated sources together. To reduce the residual signals of one source from the other separated sources (interference), we train the DNN for enhancement discriminatively to maximize the dissimilarity between the predicted sources. The experimental results show that using discriminative enhancement decreases the distortion and interference between the separated sources.

AB - The sources separated by most single channel audio source separation techniques are usually distorted and each separated source contains residual signals from the other sources. To tackle this problem, we propose to enhance the separated sources to decrease the distortion and interference between the separated sources using deep neural networks (DNNs). Two different DNNs are used in this work. The first DNN is used to separate the sources from the mixed signal. The second DNN is used to enhance the separated signals. To consider the interactions between the separated sources, we propose to use a single DNN to enhance all the separated sources together. To reduce the residual signals of one source from the other separated sources (interference), we train the DNN for enhancement discriminatively to maximize the dissimilarity between the predicted sources. The experimental results show that using discriminative enhancement decreases the distortion and interference between the separated sources.

KW - Audio enhancement

KW - Deep neural networks

KW - Discriminative training

KW - Single channel audio source separation

U2 - 10.1007/978-3-319-53547-0_23

DO - 10.1007/978-3-319-53547-0_23

M3 - Conference contribution

SN - 9783319535463

T3 - Lecture Notes in Computer Science

SP - 236

EP - 246

BT - Latent Variable Analysis and Signal Separation

A2 - Tichavský, Petr

A2 - Babaie-Zadeh, Massoud

A2 - Michel, Olivier J.J.

A2 - Thirion-Moreau, Nadège

PB - Springer, Cham

T2 - 13th International Conference on Latent Variable Analysis and Signal Separation

Y2 - 21 February 2017 through 23 February 2017

ER -

Discriminative Enhancement for Single Channel Audio Source Separation Using Deep Neural Networks

Abstract

Publication series

Conference

Access to Document

Fingerprint

Research output

Improving single-network single-channel separation of musical audio with convolutional layers

Combining mask estimates for single channel audio source separation using deep neural networks

Cite this