Reduction of Non-stationary Noise using a Non-negative Latent Variable Decomposition

Mikkel Nørgaard Schmidt; Jan Larsen

doi:10.1109/MLSP.2008.4685528

Reduction of Non-stationary Noise using a Non-negative Latent Variable Decomposition

Mikkel Nørgaard Schmidt, Jan Larsen

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

360 Downloads (Pure)

Abstract

We present a method for suppression of non-stationary noise in single channel recordings of speech. The method is based on a non-negative latent variable decomposition model for the speech and noise signals, learned directly from a noisy mixture. In non-speech regions an over complete basis is learned for the noise that is then used to jointly estimate the speech and the noise from the mixture. We compare the method to the classical spectral subtraction approach, where the noise spectrum is estimated as the average over non-speech frames. The proposed method significantly outperforms the classic approach, especially when the noise is highly non-stationary and at low signal-to-noise ratios.

Original language	English
Title of host publication	Machine Learning for Signal Processing, IEEE Workshop on
Publisher	IEEE
Publication date	2008
ISBN (Print)	978-1-4244-2375-0
DOIs	https://doi.org/10.1109/MLSP.2008.4685528
Publication status	Published - 2008
Event	2008 IEEE International Workshop on Machine Learning for Signal Processing - Cancún, Mexico Duration: 16 Oct 2008 → 19 Oct 2008 http://mlsp2008.conwiz.dk/

Workshop

Workshop	2008 IEEE International Workshop on Machine Learning for Signal Processing
Country	Mexico
City	Cancún
Period	16/10/2008 → 19/10/2008
Internet address	http://mlsp2008.conwiz.dk/

Bibliographical note

Copyright: 2008 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE

Access to Document

10.1109/MLSP.2008.4685528

SchmidtFinal published version, 581 KB

Fingerprint Dive into the research topics of 'Reduction of Non-stationary Noise using a Non-negative Latent Variable Decomposition'. Together they form a unique fingerprint.

Cite this

@inproceedings{e3a6916780104a66987c3d5e9f84d808,

title = "Reduction of Non-stationary Noise using a Non-negative Latent Variable Decomposition",

abstract = "We present a method for suppression of non-stationary noise in single channel recordings of speech. The method is based on a non-negative latent variable decomposition model for the speech and noise signals, learned directly from a noisy mixture. In non-speech regions an over complete basis is learned for the noise that is then used to jointly estimate the speech and the noise from the mixture. We compare the method to the classical spectral subtraction approach, where the noise spectrum is estimated as the average over non-speech frames. The proposed method significantly outperforms the classic approach, especially when the noise is highly non-stationary and at low signal-to-noise ratios.",

author = "Schmidt, {Mikkel N{\o}rgaard} and Jan Larsen",

note = "Copyright: 2008 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE; 2008 IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2008 ; Conference date: 16-10-2008 Through 19-10-2008",

year = "2008",

doi = "10.1109/MLSP.2008.4685528",

language = "English",

isbn = "978-1-4244-2375-0",

booktitle = "Machine Learning for Signal Processing, IEEE Workshop on",

publisher = "IEEE",

address = "United States",

url = "http://mlsp2008.conwiz.dk/",

}

TY - GEN

T1 - Reduction of Non-stationary Noise using a Non-negative Latent Variable Decomposition

AU - Schmidt, Mikkel Nørgaard

AU - Larsen, Jan

N1 - Copyright: 2008 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE

PY - 2008

Y1 - 2008

N2 - We present a method for suppression of non-stationary noise in single channel recordings of speech. The method is based on a non-negative latent variable decomposition model for the speech and noise signals, learned directly from a noisy mixture. In non-speech regions an over complete basis is learned for the noise that is then used to jointly estimate the speech and the noise from the mixture. We compare the method to the classical spectral subtraction approach, where the noise spectrum is estimated as the average over non-speech frames. The proposed method significantly outperforms the classic approach, especially when the noise is highly non-stationary and at low signal-to-noise ratios.

AB - We present a method for suppression of non-stationary noise in single channel recordings of speech. The method is based on a non-negative latent variable decomposition model for the speech and noise signals, learned directly from a noisy mixture. In non-speech regions an over complete basis is learned for the noise that is then used to jointly estimate the speech and the noise from the mixture. We compare the method to the classical spectral subtraction approach, where the noise spectrum is estimated as the average over non-speech frames. The proposed method significantly outperforms the classic approach, especially when the noise is highly non-stationary and at low signal-to-noise ratios.

U2 - 10.1109/MLSP.2008.4685528

DO - 10.1109/MLSP.2008.4685528

M3 - Article in proceedings

SN - 978-1-4244-2375-0

BT - Machine Learning for Signal Processing, IEEE Workshop on

PB - IEEE

T2 - 2008 IEEE International Workshop on Machine Learning for Signal Processing

Y2 - 16 October 2008 through 19 October 2008

ER -