Examining The Perceptual Effect of Alternative Objective Functions for Deep Learning Based Music Source Separation

Stylianos Ioannis Mimilakis; Estefania Cano; Derry FitzGerald; Konstantinos Drossos; Gerald Schuller

doi:10.1109/ACSSC.2018.8645257

Examining The Perceptual Effect of Alternative Objective Functions for Deep Learning Based Music Source Separation

Stylianos Ioannis Mimilakis
, Estefania Cano
, Derry FitzGerald
, Konstantinos Drossos
, Gerald Schuller

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

2 Citations (Scopus)

Abstract

In this study, we examine the effect of various objective functions used to optimize the recently proposed deep learning architecture for singing voice separation MaD - Masker and Denoiser. The parameters of the MaD architecture are optimized using an objective function that contains a reconstruction criterion between predicted and true magnitude spectra of the singing voice, and a regularization term. We examine various reconstruction criteria such as the generalized Kullback-Leibler, mean squared error, and noise to mask ratio. We also explore recently proposed, for optimizing MaD, regularization terms such as sparsity and TwinNetwork regularization. Results from both objective assessment and listening tests suggest that the TwinNetwork regularization results in improved singing voice separation quality.

Original language	English
Title of host publication	2018 52nd Asilomar Conference on Signals, Systems, and Computers
Publisher	IEEE
Pages	679-683
ISBN (Electronic)	978-1-5386-9218-9
ISBN (Print)	978-1-5386-9219-6
DOIs	https://doi.org/10.1109/ACSSC.2018.8645257
Publication status	Published - Oct 2018
Publication type	A4 Article in conference proceedings
Event	Asilomar Conference on Signals, Systems and Computers - Duration: 1 Jan 1900 → …

Publication series

Name
ISSN (Print)	1058-6393

Conference

Conference	Asilomar Conference on Signals, Systems and Computers
Period	1/01/00 → …

Publication forum classification

Publication forum level 1

Access to Document

10.1109/ACSSC.2018.8645257

Support Material for: S.I. Mimilakis et al., ``Examining the Perceptual Effect of Alternative Objective Functions for Deep Learning Based Music Source Separation''
Mimilakis, S. I. (Creator), Drossos, K. (Creator), Cano, E. (Creator) & Schuller, G. (Creator), Zenodo, 2 Nov 2018
DOI: 10.5281/zenodo.1476866
Dataset

Cite this

@inproceedings{f7ba4aa414074d0ab37bc30eedac3224,

title = "Examining The Perceptual Effect of Alternative Objective Functions for Deep Learning Based Music Source Separation",

abstract = "In this study, we examine the effect of various objective functions used to optimize the recently proposed deep learning architecture for singing voice separation MaD - Masker and Denoiser. The parameters of the MaD architecture are optimized using an objective function that contains a reconstruction criterion between predicted and true magnitude spectra of the singing voice, and a regularization term. We examine various reconstruction criteria such as the generalized Kullback-Leibler, mean squared error, and noise to mask ratio. We also explore recently proposed, for optimizing MaD, regularization terms such as sparsity and TwinNetwork regularization. Results from both objective assessment and listening tests suggest that the TwinNetwork regularization results in improved singing voice separation quality.",

author = "Mimilakis, \{Stylianos Ioannis\} and Estefania Cano and Derry FitzGerald and Konstantinos Drossos and Gerald Schuller",

year = "2018",

month = oct,

doi = "10.1109/ACSSC.2018.8645257",

language = "English",

isbn = "978-1-5386-9219-6",

publisher = "IEEE",

pages = "679--683",

booktitle = "2018 52nd Asilomar Conference on Signals, Systems, and Computers",

address = "United States",

note = "Asilomar Conference on Signals, Systems and Computers ; Conference date: 01-01-1900",

}

Mimilakis, SI, Cano, E, FitzGerald, D, Drossos, K & Schuller, G 2018, Examining The Perceptual Effect of Alternative Objective Functions for Deep Learning Based Music Source Separation. in 2018 52nd Asilomar Conference on Signals, Systems, and Computers. IEEE, pp. 679-683, Asilomar Conference on Signals, Systems and Computers, 1/01/00. https://doi.org/10.1109/ACSSC.2018.8645257

Examining The Perceptual Effect of Alternative Objective Functions for Deep Learning Based Music Source Separation. / Mimilakis, Stylianos Ioannis; Cano, Estefania; FitzGerald, Derry et al.
2018 52nd Asilomar Conference on Signals, Systems, and Computers. IEEE, 2018. p. 679-683.

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

TY - GEN

T1 - Examining The Perceptual Effect of Alternative Objective Functions for Deep Learning Based Music Source Separation

AU - Mimilakis, Stylianos Ioannis

AU - Cano, Estefania

AU - FitzGerald, Derry

AU - Drossos, Konstantinos

AU - Schuller, Gerald

PY - 2018/10

Y1 - 2018/10

N2 - In this study, we examine the effect of various objective functions used to optimize the recently proposed deep learning architecture for singing voice separation MaD - Masker and Denoiser. The parameters of the MaD architecture are optimized using an objective function that contains a reconstruction criterion between predicted and true magnitude spectra of the singing voice, and a regularization term. We examine various reconstruction criteria such as the generalized Kullback-Leibler, mean squared error, and noise to mask ratio. We also explore recently proposed, for optimizing MaD, regularization terms such as sparsity and TwinNetwork regularization. Results from both objective assessment and listening tests suggest that the TwinNetwork regularization results in improved singing voice separation quality.

AB - In this study, we examine the effect of various objective functions used to optimize the recently proposed deep learning architecture for singing voice separation MaD - Masker and Denoiser. The parameters of the MaD architecture are optimized using an objective function that contains a reconstruction criterion between predicted and true magnitude spectra of the singing voice, and a regularization term. We examine various reconstruction criteria such as the generalized Kullback-Leibler, mean squared error, and noise to mask ratio. We also explore recently proposed, for optimizing MaD, regularization terms such as sparsity and TwinNetwork regularization. Results from both objective assessment and listening tests suggest that the TwinNetwork regularization results in improved singing voice separation quality.

U2 - 10.1109/ACSSC.2018.8645257

DO - 10.1109/ACSSC.2018.8645257

M3 - Conference contribution

SN - 978-1-5386-9219-6

SP - 679

EP - 683

BT - 2018 52nd Asilomar Conference on Signals, Systems, and Computers

PB - IEEE

T2 - Asilomar Conference on Signals, Systems and Computers

Y2 - 1 January 1900

ER -

Examining The Perceptual Effect of Alternative Objective Functions for Deep Learning Based Music Source Separation

Abstract

Publication series

Conference

Publication forum classification

Access to Document

Fingerprint

Datasets

Support Material for: S.I. Mimilakis et al., ``Examining the Perceptual Effect of Alternative Objective Functions for Deep Learning Based Music Source Separation''

Cite this