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

Andrew JR Simpson; Gerard Roma; Mark D Plumbley

doi:10.1007/978-3-319-22482-4_50

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

Andrew JR Simpson, Gerard Roma, Mark D Plumbley

University of Surrey

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

72 Citations (Scopus)

Abstract

Identification and extraction of singing voice from within musical mixtures is a key challenge in source separation and machine audition. Recently, deep neural networks (DNN) have been used to estimate ‘ideal’ binary masks for carefully controlled cocktail party speech separation problems. However, it is not yet known whether these methods are capable of generalizing to the discrimination of voice and non-voice in the context of musical mixtures. Here, we trained a convolutional DNN (of around a billion parameters) to provide probabilistic estimates of the ideal binary mask for separation of vocal sounds from real-world musical mixtures. We contrast our DNN results with more traditional linear methods. Our approach may be useful for automatic removal of vocal sounds from musical mixtures for ‘karaoke’ type applications.

Original language	English
Title of host publication	Latent Variable Analysis and Signal Separation
Editors	Emmanuel Vincent, Arie Yeredor, Zbyněk Koldovský, Petr Tichavský
Publisher	Springer, Cham
Pages	429-436
Number of pages	8
ISBN (Electronic)	9783319224824
ISBN (Print)	9783319224817
DOIs	https://doi.org/10.1007/978-3-319-22482-4_50
Publication status	Published - 15 Aug 2015
Externally published	Yes
Event	12th International Conference on Latent Variable Analysis and Signal Separation - Technical University of Liberec, Liberec, Czech Republic Duration: 25 Aug 2015 → 28 Aug 2015 Conference number: 12

Publication series

Name	Lecture Notes in Computer Science
Publisher	Springer
Volume	9237
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	12th International Conference on Latent Variable Analysis and Signal Separation
Abbreviated title	LVA/ICA 2015
Country/Territory	Czech Republic
City	Liberec
Period	25/08/15 → 28/08/15

Access to Document

10.1007/978-3-319-22482-4_50Licence: 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

Simpson, A. JR., Roma, G., & Plumbley, M. D. (2015). Deep Karaoke: Extracting Vocals from Musical Mixtures Using a Convolutional Deep Neural Network. In E. Vincent, A. Yeredor, Z. Koldovský, & P. Tichavský (Eds.), Latent Variable Analysis and Signal Separation (pp. 429-436). (Lecture Notes in Computer Science; Vol. 9237). Springer, Cham. https://doi.org/10.1007/978-3-319-22482-4_50

@inproceedings{4c378b1010ab4456988ce6ee1f333c42,

title = "Deep Karaoke: Extracting Vocals from Musical Mixtures Using a Convolutional Deep Neural Network",

abstract = "Identification and extraction of singing voice from within musical mixtures is a key challenge in source separation and machine audition. Recently, deep neural networks (DNN) have been used to estimate {\textquoteleft}ideal{\textquoteright} binary masks for carefully controlled cocktail party speech separation problems. However, it is not yet known whether these methods are capable of generalizing to the discrimination of voice and non-voice in the context of musical mixtures. Here, we trained a convolutional DNN (of around a billion parameters) to provide probabilistic estimates of the ideal binary mask for separation of vocal sounds from real-world musical mixtures. We contrast our DNN results with more traditional linear methods. Our approach may be useful for automatic removal of vocal sounds from musical mixtures for {\textquoteleft}karaoke{\textquoteright} type applications.",

keywords = "Convolution, Deep learning, Source separation, Supervised learning",

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

year = "2015",

month = aug,

day = "15",

doi = "10.1007/978-3-319-22482-4\_50",

language = "English",

isbn = "9783319224817",

series = "Lecture Notes in Computer Science",

publisher = "Springer, Cham",

pages = "429--436",

editor = "Emmanuel Vincent and Arie Yeredor and Zbyn{\v e}k Koldovsk{\'y} and Petr Tichavsk{\'y}",

booktitle = "Latent Variable Analysis and Signal Separation",

address = "Switzerland",

note = "12th International Conference on Latent Variable Analysis and Signal Separation, LVA/ICA 2015 ; Conference date: 25-08-2015 Through 28-08-2015",

}

Simpson, AJR, Roma, G & Plumbley, MD 2015, Deep Karaoke: Extracting Vocals from Musical Mixtures Using a Convolutional Deep Neural Network. in E Vincent, A Yeredor, Z Koldovský & P Tichavský (eds), Latent Variable Analysis and Signal Separation. Lecture Notes in Computer Science, vol. 9237, Springer, Cham, pp. 429-436, 12th International Conference on Latent Variable Analysis and Signal Separation, Liberec, Czech Republic, 25/08/15. https://doi.org/10.1007/978-3-319-22482-4_50

Deep Karaoke: Extracting Vocals from Musical Mixtures Using a Convolutional Deep Neural Network. / Simpson, Andrew JR; Roma, Gerard; Plumbley, Mark D.
Latent Variable Analysis and Signal Separation. ed. / Emmanuel Vincent; Arie Yeredor; Zbyněk Koldovský; Petr Tichavský. Springer, Cham, 2015. p. 429-436 (Lecture Notes in Computer Science; Vol. 9237).

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

TY - GEN

T1 - Deep Karaoke

T2 - 12th International Conference on Latent Variable Analysis and Signal Separation

AU - Simpson, Andrew JR

AU - Roma, Gerard

AU - Plumbley, Mark D

N1 - Conference code: 12

PY - 2015/8/15

Y1 - 2015/8/15

N2 - Identification and extraction of singing voice from within musical mixtures is a key challenge in source separation and machine audition. Recently, deep neural networks (DNN) have been used to estimate ‘ideal’ binary masks for carefully controlled cocktail party speech separation problems. However, it is not yet known whether these methods are capable of generalizing to the discrimination of voice and non-voice in the context of musical mixtures. Here, we trained a convolutional DNN (of around a billion parameters) to provide probabilistic estimates of the ideal binary mask for separation of vocal sounds from real-world musical mixtures. We contrast our DNN results with more traditional linear methods. Our approach may be useful for automatic removal of vocal sounds from musical mixtures for ‘karaoke’ type applications.

AB - Identification and extraction of singing voice from within musical mixtures is a key challenge in source separation and machine audition. Recently, deep neural networks (DNN) have been used to estimate ‘ideal’ binary masks for carefully controlled cocktail party speech separation problems. However, it is not yet known whether these methods are capable of generalizing to the discrimination of voice and non-voice in the context of musical mixtures. Here, we trained a convolutional DNN (of around a billion parameters) to provide probabilistic estimates of the ideal binary mask for separation of vocal sounds from real-world musical mixtures. We contrast our DNN results with more traditional linear methods. Our approach may be useful for automatic removal of vocal sounds from musical mixtures for ‘karaoke’ type applications.

KW - Convolution

KW - Deep learning

KW - Source separation

KW - Supervised learning

U2 - 10.1007/978-3-319-22482-4_50

DO - 10.1007/978-3-319-22482-4_50

M3 - Conference contribution

SN - 9783319224817

T3 - Lecture Notes in Computer Science

SP - 429

EP - 436

BT - Latent Variable Analysis and Signal Separation

A2 - Vincent, Emmanuel

A2 - Yeredor, Arie

A2 - Koldovský, Zbyněk

A2 - Tichavský, Petr

PB - Springer, Cham

Y2 - 25 August 2015 through 28 August 2015

ER -

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

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