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

Andrew JR Simpson, Gerard Roma, Mark D Plumbley

Research output: Chapter in Book/Report/Conference proceedingConference contribution

29 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.
LanguageEnglish
Title of host publicationLatent Variable Analysis and Signal Separation
EditorsEmmanuel Vincent, Arie Yeredor, Zbyněk Koldovský, Petr Tichavský
PublisherSpringer, Cham
Pages429-436
Number of pages8
ISBN (Electronic)9783319224824
ISBN (Print)9783319224817
DOIs
Publication statusPublished - 15 Aug 2015
Externally publishedYes
Event12th International Conference on Latent Variable Analysis and Signal Separation - Technical University of Liberec, Liberec, Czech Republic
Duration: 25 Aug 201528 Aug 2015
Conference number: 12

Publication series

NameLecture Notes in Computer Science
PublisherSpringer
Volume9237
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference12th International Conference on Latent Variable Analysis and Signal Separation
Abbreviated titleLVA/ICA 2015
CountryCzech Republic
CityLiberec
Period25/08/1528/08/15

Fingerprint

Masks
Acoustic waves
Source separation
Audition
Deep neural networks

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
Simpson, Andrew JR ; Roma, Gerard ; Plumbley, Mark D. / Deep Karaoke : Extracting Vocals from Musical Mixtures Using a Convolutional Deep Neural Network. Latent Variable Analysis and Signal Separation. editor / Emmanuel Vincent ; Arie Yeredor ; Zbyněk Koldovský ; Petr Tichavský. Springer, Cham, 2015. pp. 429-436 (Lecture Notes in Computer Science).
@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 ‘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.",
keywords = "Convolution, Deep learning, Source separation, Supervised learning",
author = "Simpson, {Andrew JR} and Gerard Roma and Plumbley, {Mark D}",
year = "2015",
month = "8",
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 Zbyněk Koldovsk{\'y} and Petr Tichavsk{\'y}",
booktitle = "Latent Variable Analysis and Signal Separation",

}

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 proceedingConference contribution

TY - GEN

T1 - Deep Karaoke

T2 - Extracting Vocals from Musical Mixtures Using a Convolutional Deep Neural Network

AU - Simpson, Andrew JR

AU - Roma, Gerard

AU - Plumbley, Mark D

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

ER -

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