A new log-spectral amplitude estimator using the weighted Euclidean distortion measure for speech enhancement

Anh Duc Nguyen, Kensuke Naoe, Yoshiyasu Takefuji

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

1 Citation (Scopus)

Abstract

This paper considers a perceptually motivated estimator for single-channel speech enhancement based on statistics and Bayesian estimation in the frequency domain. In detail, we propose a new speech log-spectral amplitude estimator where the cost function is the weighted Euclidean distortion measure of the speech log-spectral amplitude. That cost function is motivated by auditory masking effects of the human hearing system. The statistical assumptions used to develop the proposed estimator are the complex Gaussian distribution and independence of speech, noise discrete Fourier transform coefficients. We evaluate this estimator with speech signals contaminated by various noise sources at different input signal-to-noise ratios and find that it achieves better performance than the well-known minimum mean square error log-spectral amplitude estimator in terms of both noise reduction and speech quality.

Original languageEnglish
Title of host publication2010 IEEE 26th Convention of Electrical and Electronics Engineers in Israel, IEEEI 2010
Pages675-679
Number of pages5
DOIs
Publication statusPublished - 2010
Externally publishedYes
Event2010 IEEE 26th Convention of Electrical and Electronics Engineers in Israel, IEEEI 2010 - Eilat, Israel
Duration: 2010 Nov 172010 Nov 20

Publication series

Name2010 IEEE 26th Convention of Electrical and Electronics Engineers in Israel, IEEEI 2010

Other

Other2010 IEEE 26th Convention of Electrical and Electronics Engineers in Israel, IEEEI 2010
Country/TerritoryIsrael
CityEilat
Period10/11/1710/11/20

Keywords

  • Bayesian estimation
  • Noise reduction
  • Short-time spectral amplitude
  • Speech enhancement
  • Weighted Euclidean distortion measure

ASJC Scopus subject areas

  • Electrical and Electronic Engineering

Fingerprint

Dive into the research topics of 'A new log-spectral amplitude estimator using the weighted Euclidean distortion measure for speech enhancement'. Together they form a unique fingerprint.

Cite this