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

Gerard Roma; Owen Green; Pierre Alexandre Tremblay

doi:10.1007/978-3-319-93764-9_29

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

Gerard Roma, Owen Green, Pierre Alexandre Tremblay

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

6 Citations (Scopus)

Abstract

Most convolutional neural network architectures explored so far for musical audio separation follow an autoencoder structure, where the mixture is considered to be a corrupted version of the original source. On the other hand, many approaches based on deep neural networks make use of several networks with different objectives for estimating the sources. In this paper we propose a discriminative approach based on traditional convolutional neural network architectures for image classification and speech recognition. Our results show that this architecture performs similarly to current state of the art approaches for separating singing voice, and that the addition of convolutional layers allows improving separation results with respect to using only fully-connected layers.

Original language	English
Title of host publication	Latent Variable Analysis and Signal Separation
Subtitle of host publication	14th International Conference, LVA/ICA 2018, Guildford, UK, July 2–5, 2018, Proceedings
Editors	Sharon Gannot, Yannick Deville, Russell Mason, Mark D. Plumbley, Dominic Ward
Publisher	Springer Verlag
Pages	306-315
ISBN (Electronic)	9783319937649
ISBN (Print)	9783319937632
DOIs	https://doi.org/10.1007/978-3-319-93764-9_29
Publication status	Published - 6 Jun 2018
Event	14th International Conference on Latent Variable Analysis and Signal Seperation - University of Surrey, Guildford, United Kingdom Duration: 2 Jul 2018 → 6 Jul 2018 http://cvssp.org/events/lva-ica-2018/ (Link to Conference Website)

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	10891 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	14th International Conference on Latent Variable Analysis and Signal Seperation
Abbreviated title	LVA / ICA 2018
Country/Territory	United Kingdom
City	Guildford
Period	2/07/18 → 6/07/18
Internet address	http://cvssp.org/events/lva-ica-2018/ (Link to Conference Website)

Access to Document

10.1007/978-3-319-93764-9_29Licence: Unspecified

lva_ica_057_revisedAccepted author manuscript, 278 KBLicence: Unspecified

Cite this

Roma, G., Green, O., & Tremblay, P. A. (2018). Improving single-network single-channel separation of musical audio with convolutional layers. In S. Gannot, Y. Deville, R. Mason, M. D. Plumbley, & D. Ward (Eds.), Latent Variable Analysis and Signal Separation: 14th International Conference, LVA/ICA 2018, Guildford, UK, July 2–5, 2018, Proceedings (pp. 306-315). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 10891 LNCS). Springer Verlag. https://doi.org/10.1007/978-3-319-93764-9_29

Roma, Gerard ; Green, Owen ; Tremblay, Pierre Alexandre. / Improving single-network single-channel separation of musical audio with convolutional layers. Latent Variable Analysis and Signal Separation: 14th International Conference, LVA/ICA 2018, Guildford, UK, July 2–5, 2018, Proceedings. editor / Sharon Gannot ; Yannick Deville ; Russell Mason ; Mark D. Plumbley ; Dominic Ward. Springer Verlag, 2018. pp. 306-315 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{e6448cb4ea924a7496cbd1ab92fd4b4a,

title = "Improving single-network single-channel separation of musical audio with convolutional layers",

abstract = "Most convolutional neural network architectures explored so far for musical audio separation follow an autoencoder structure, where the mixture is considered to be a corrupted version of the original source. On the other hand, many approaches based on deep neural networks make use of several networks with different objectives for estimating the sources. In this paper we propose a discriminative approach based on traditional convolutional neural network architectures for image classification and speech recognition. Our results show that this architecture performs similarly to current state of the art approaches for separating singing voice, and that the addition of convolutional layers allows improving separation results with respect to using only fully-connected layers.",

keywords = "audio source separation, convolutional neural networks, Audio source separation, Convolutional neural networks",

author = "Gerard Roma and Owen Green and Tremblay, \{Pierre Alexandre\}",

note = "The final authenticated publication is available online at https://doi.org/10.1007/978-3-319-93764-9\_29; 14th International Conference on Latent Variable Analysis and Signal Seperation, LVA / ICA 2018 ; Conference date: 02-07-2018 Through 06-07-2018",

year = "2018",

month = jun,

day = "6",

doi = "10.1007/978-3-319-93764-9\_29",

language = "English",

isbn = "9783319937632",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer Verlag",

pages = "306--315",

editor = "Sharon Gannot and Yannick Deville and Russell Mason and Plumbley, \{Mark D.\} and Dominic Ward",

booktitle = "Latent Variable Analysis and Signal Separation",

address = "Germany",

url = "http://cvssp.org/events/lva-ica-2018/",

}

Roma, G, Green, O & Tremblay, PA 2018, Improving single-network single-channel separation of musical audio with convolutional layers. in S Gannot, Y Deville, R Mason, MD Plumbley & D Ward (eds), Latent Variable Analysis and Signal Separation: 14th International Conference, LVA/ICA 2018, Guildford, UK, July 2–5, 2018, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 10891 LNCS, Springer Verlag, pp. 306-315, 14th International Conference on Latent Variable Analysis and Signal Seperation, Guildford, United Kingdom, 2/07/18. https://doi.org/10.1007/978-3-319-93764-9_29

Improving single-network single-channel separation of musical audio with convolutional layers. / Roma, Gerard; Green, Owen; Tremblay, Pierre Alexandre.
Latent Variable Analysis and Signal Separation: 14th International Conference, LVA/ICA 2018, Guildford, UK, July 2–5, 2018, Proceedings. ed. / Sharon Gannot; Yannick Deville; Russell Mason; Mark D. Plumbley; Dominic Ward. Springer Verlag, 2018. 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

TY - GEN

T1 - Improving single-network single-channel separation of musical audio with convolutional layers

AU - Roma, Gerard

AU - Green, Owen

AU - Tremblay, Pierre Alexandre

N1 - The final authenticated publication is available online at https://doi.org/10.1007/978-3-319-93764-9_29

PY - 2018/6/6

Y1 - 2018/6/6

N2 - Most convolutional neural network architectures explored so far for musical audio separation follow an autoencoder structure, where the mixture is considered to be a corrupted version of the original source. On the other hand, many approaches based on deep neural networks make use of several networks with different objectives for estimating the sources. In this paper we propose a discriminative approach based on traditional convolutional neural network architectures for image classification and speech recognition. Our results show that this architecture performs similarly to current state of the art approaches for separating singing voice, and that the addition of convolutional layers allows improving separation results with respect to using only fully-connected layers.

AB - Most convolutional neural network architectures explored so far for musical audio separation follow an autoencoder structure, where the mixture is considered to be a corrupted version of the original source. On the other hand, many approaches based on deep neural networks make use of several networks with different objectives for estimating the sources. In this paper we propose a discriminative approach based on traditional convolutional neural network architectures for image classification and speech recognition. Our results show that this architecture performs similarly to current state of the art approaches for separating singing voice, and that the addition of convolutional layers allows improving separation results with respect to using only fully-connected layers.

KW - audio source separation

KW - convolutional neural networks

KW - Audio source separation

KW - Convolutional neural networks

UR - http://cvssp.org/events/lva-ica-2018/

UR - https://www.scopus.com/pages/publications/85048585573

U2 - 10.1007/978-3-319-93764-9_29

DO - 10.1007/978-3-319-93764-9_29

M3 - Conference contribution

SN - 9783319937632

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 306

EP - 315

BT - Latent Variable Analysis and Signal Separation

A2 - Gannot, Sharon

A2 - Deville, Yannick

A2 - Mason, Russell

A2 - Plumbley, Mark D.

A2 - Ward, Dominic

PB - Springer Verlag

T2 - 14th International Conference on Latent Variable Analysis and Signal Seperation

Y2 - 2 July 2018 through 6 July 2018

ER -

Roma G, Green O, Tremblay PA. Improving single-network single-channel separation of musical audio with convolutional layers. In Gannot S, Deville Y, Mason R, Plumbley MD, Ward D, editors, Latent Variable Analysis and Signal Separation: 14th International Conference, LVA/ICA 2018, Guildford, UK, July 2–5, 2018, Proceedings. Springer Verlag. 2018. p. 306-315. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-319-93764-9_29

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

Abstract

Publication series

Conference

Access to Document

Other files and links

Fingerprint

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

Two-stage single-channel audio source separation using deep neural networks

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

Single-Channel Audio Source Separation Using Deep Neural Network Ensembles

Deep Karaoke: Extracting Vocals from Musical Mixtures Using a Convolutional Deep Neural Network

FluCoMa: Fluid Corpus Manipulations

Cite this

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

Abstract

Publication series

Conference

Access to Document

Other files and links

Fingerprint

Research output

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

Two-stage single-channel audio source separation using deep neural networks

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

Single-Channel Audio Source Separation Using Deep Neural Network Ensembles

Deep Karaoke: Extracting Vocals from Musical Mixtures Using a Convolutional Deep Neural Network

Projects

FluCoMa: Fluid Corpus Manipulations

Cite this