Source separation with scattering Non-Negative Matrix Factorization

Joan Bruna Estrach, Pablo Sprechmann, Yann LeCun

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

Abstract

This paper presents a single-channel source separation method that extends the ideas of Nonnegative Matrix Factorization (NMF). We interpret the approach of audio demixing via NMF as a cascade of a pooled analysis operator, given for example by the magnitude spectrogram, and a synthesis operators given by the matrix decomposition. Instead of imposing the temporal consistency of the decomposition through sophisticated structured penalties in the synthesis stage, we propose to change the analysis operator for a deep scattering representation, where signals are represented at several time resolutions. This new signal representation is invariant to smooth changes in the signal, consistent with its temporal dynamics. We evaluate the proposed approach in a speech separation task obtaining promising results.

Original languageEnglish (US)
Title of host publication2015 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2015 - Proceedings
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages1876-1880
Number of pages5
Volume2015-August
ISBN (Print)9781467369978
DOIs
StatePublished - Aug 4 2015
Event40th IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2015 - Brisbane, Australia
Duration: Apr 19 2014Apr 24 2014

Other

Other40th IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2015
CountryAustralia
CityBrisbane
Period4/19/144/24/14

Fingerprint

Source separation
Factorization
Scattering
Decomposition
Mathematical operators

Keywords

  • non-negative matrix factorization
  • scattering
  • source separation

ASJC Scopus subject areas

  • Signal Processing
  • Software
  • Electrical and Electronic Engineering

Cite this

Bruna Estrach, J., Sprechmann, P., & LeCun, Y. (2015). Source separation with scattering Non-Negative Matrix Factorization. In 2015 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2015 - Proceedings (Vol. 2015-August, pp. 1876-1880). [7178296] Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/ICASSP.2015.7178296

Source separation with scattering Non-Negative Matrix Factorization. / Bruna Estrach, Joan; Sprechmann, Pablo; LeCun, Yann.

2015 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2015 - Proceedings. Vol. 2015-August Institute of Electrical and Electronics Engineers Inc., 2015. p. 1876-1880 7178296.

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

Bruna Estrach, J, Sprechmann, P & LeCun, Y 2015, Source separation with scattering Non-Negative Matrix Factorization. in 2015 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2015 - Proceedings. vol. 2015-August, 7178296, Institute of Electrical and Electronics Engineers Inc., pp. 1876-1880, 40th IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2015, Brisbane, Australia, 4/19/14. https://doi.org/10.1109/ICASSP.2015.7178296
Bruna Estrach J, Sprechmann P, LeCun Y. Source separation with scattering Non-Negative Matrix Factorization. In 2015 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2015 - Proceedings. Vol. 2015-August. Institute of Electrical and Electronics Engineers Inc. 2015. p. 1876-1880. 7178296 https://doi.org/10.1109/ICASSP.2015.7178296
Bruna Estrach, Joan ; Sprechmann, Pablo ; LeCun, Yann. / Source separation with scattering Non-Negative Matrix Factorization. 2015 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2015 - Proceedings. Vol. 2015-August Institute of Electrical and Electronics Engineers Inc., 2015. pp. 1876-1880
@inproceedings{fe7d85e3403148bebf5c72d59b7380d2,
title = "Source separation with scattering Non-Negative Matrix Factorization",
abstract = "This paper presents a single-channel source separation method that extends the ideas of Nonnegative Matrix Factorization (NMF). We interpret the approach of audio demixing via NMF as a cascade of a pooled analysis operator, given for example by the magnitude spectrogram, and a synthesis operators given by the matrix decomposition. Instead of imposing the temporal consistency of the decomposition through sophisticated structured penalties in the synthesis stage, we propose to change the analysis operator for a deep scattering representation, where signals are represented at several time resolutions. This new signal representation is invariant to smooth changes in the signal, consistent with its temporal dynamics. We evaluate the proposed approach in a speech separation task obtaining promising results.",
keywords = "non-negative matrix factorization, scattering, source separation",
author = "{Bruna Estrach}, Joan and Pablo Sprechmann and Yann LeCun",
year = "2015",
month = "8",
day = "4",
doi = "10.1109/ICASSP.2015.7178296",
language = "English (US)",
isbn = "9781467369978",
volume = "2015-August",
pages = "1876--1880",
booktitle = "2015 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2015 - Proceedings",
publisher = "Institute of Electrical and Electronics Engineers Inc.",

}

TY - GEN

T1 - Source separation with scattering Non-Negative Matrix Factorization

AU - Bruna Estrach, Joan

AU - Sprechmann, Pablo

AU - LeCun, Yann

PY - 2015/8/4

Y1 - 2015/8/4

N2 - This paper presents a single-channel source separation method that extends the ideas of Nonnegative Matrix Factorization (NMF). We interpret the approach of audio demixing via NMF as a cascade of a pooled analysis operator, given for example by the magnitude spectrogram, and a synthesis operators given by the matrix decomposition. Instead of imposing the temporal consistency of the decomposition through sophisticated structured penalties in the synthesis stage, we propose to change the analysis operator for a deep scattering representation, where signals are represented at several time resolutions. This new signal representation is invariant to smooth changes in the signal, consistent with its temporal dynamics. We evaluate the proposed approach in a speech separation task obtaining promising results.

AB - This paper presents a single-channel source separation method that extends the ideas of Nonnegative Matrix Factorization (NMF). We interpret the approach of audio demixing via NMF as a cascade of a pooled analysis operator, given for example by the magnitude spectrogram, and a synthesis operators given by the matrix decomposition. Instead of imposing the temporal consistency of the decomposition through sophisticated structured penalties in the synthesis stage, we propose to change the analysis operator for a deep scattering representation, where signals are represented at several time resolutions. This new signal representation is invariant to smooth changes in the signal, consistent with its temporal dynamics. We evaluate the proposed approach in a speech separation task obtaining promising results.

KW - non-negative matrix factorization

KW - scattering

KW - source separation

UR - http://www.scopus.com/inward/record.url?scp=84941339663&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84941339663&partnerID=8YFLogxK

U2 - 10.1109/ICASSP.2015.7178296

DO - 10.1109/ICASSP.2015.7178296

M3 - Conference contribution

SN - 9781467369978

VL - 2015-August

SP - 1876

EP - 1880

BT - 2015 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2015 - Proceedings

PB - Institute of Electrical and Electronics Engineers Inc.

ER -