Single-channel speech enhancement method using reconstructive NMF with spectrotemporal speech presence probabilities

Seongjae Lee, David K. Han, Hanseok Ko

Research output: Contribution to journalArticle

10 Citations (Scopus)

Abstract

In this paper, a novel single microphone channel-based speech enhancement technique is presented. While most of the conventional nonnegative matrix factorization-based approaches focus on generating a basis matrix of speech and noise for enhancement, the proposed algorithm performs an additional process to reconstruct speech from noisy speech when these two elements are highly overlapped in selected spectral bands. This process involves a log-spectral amplitude based estimator, which provides the spectrotemporal speech presence probability to obtain a more accurate reconstruction. Moreover, the proposed algorithm applies an unsupervised learning method to the input noise, so it is adaptable to any type of environmental noise without a pre-trained dictionary. The experimental results demonstrate that the proposed algorithm obtains improved speech enhancement performance compared with conventional single channel-based approaches.

Original languageEnglish
JournalApplied Acoustics
DOIs
Publication statusAccepted/In press - 2015 Aug 31

Fingerprint

augmentation
dictionaries
spectral bands
matrices
microphones
factorization
estimators
learning

Keywords

  • Acoustic source separation
  • Noise reduction
  • Speech enhancement

ASJC Scopus subject areas

  • Acoustics and Ultrasonics

Cite this

Single-channel speech enhancement method using reconstructive NMF with spectrotemporal speech presence probabilities. / Lee, Seongjae; Han, David K.; Ko, Hanseok.

In: Applied Acoustics, 31.08.2015.

Research output: Contribution to journalArticle

@article{84d042126c8b4fbbba785bc72ca587db,
title = "Single-channel speech enhancement method using reconstructive NMF with spectrotemporal speech presence probabilities",
abstract = "In this paper, a novel single microphone channel-based speech enhancement technique is presented. While most of the conventional nonnegative matrix factorization-based approaches focus on generating a basis matrix of speech and noise for enhancement, the proposed algorithm performs an additional process to reconstruct speech from noisy speech when these two elements are highly overlapped in selected spectral bands. This process involves a log-spectral amplitude based estimator, which provides the spectrotemporal speech presence probability to obtain a more accurate reconstruction. Moreover, the proposed algorithm applies an unsupervised learning method to the input noise, so it is adaptable to any type of environmental noise without a pre-trained dictionary. The experimental results demonstrate that the proposed algorithm obtains improved speech enhancement performance compared with conventional single channel-based approaches.",
keywords = "Acoustic source separation, Noise reduction, Speech enhancement",
author = "Seongjae Lee and Han, {David K.} and Hanseok Ko",
year = "2015",
month = "8",
day = "31",
doi = "10.1016/j.apacoust.2016.04.024",
language = "English",
journal = "Applied Acoustics",
issn = "0003-682X",
publisher = "Elsevier Limited",

}

TY - JOUR

T1 - Single-channel speech enhancement method using reconstructive NMF with spectrotemporal speech presence probabilities

AU - Lee, Seongjae

AU - Han, David K.

AU - Ko, Hanseok

PY - 2015/8/31

Y1 - 2015/8/31

N2 - In this paper, a novel single microphone channel-based speech enhancement technique is presented. While most of the conventional nonnegative matrix factorization-based approaches focus on generating a basis matrix of speech and noise for enhancement, the proposed algorithm performs an additional process to reconstruct speech from noisy speech when these two elements are highly overlapped in selected spectral bands. This process involves a log-spectral amplitude based estimator, which provides the spectrotemporal speech presence probability to obtain a more accurate reconstruction. Moreover, the proposed algorithm applies an unsupervised learning method to the input noise, so it is adaptable to any type of environmental noise without a pre-trained dictionary. The experimental results demonstrate that the proposed algorithm obtains improved speech enhancement performance compared with conventional single channel-based approaches.

AB - In this paper, a novel single microphone channel-based speech enhancement technique is presented. While most of the conventional nonnegative matrix factorization-based approaches focus on generating a basis matrix of speech and noise for enhancement, the proposed algorithm performs an additional process to reconstruct speech from noisy speech when these two elements are highly overlapped in selected spectral bands. This process involves a log-spectral amplitude based estimator, which provides the spectrotemporal speech presence probability to obtain a more accurate reconstruction. Moreover, the proposed algorithm applies an unsupervised learning method to the input noise, so it is adaptable to any type of environmental noise without a pre-trained dictionary. The experimental results demonstrate that the proposed algorithm obtains improved speech enhancement performance compared with conventional single channel-based approaches.

KW - Acoustic source separation

KW - Noise reduction

KW - Speech enhancement

UR - http://www.scopus.com/inward/record.url?scp=84970039128&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84970039128&partnerID=8YFLogxK

U2 - 10.1016/j.apacoust.2016.04.024

DO - 10.1016/j.apacoust.2016.04.024

M3 - Article

JO - Applied Acoustics

JF - Applied Acoustics

SN - 0003-682X

ER -