A perceptually motivated approach for speech enhancement

被引：60

作者：

Hu, Y ^{[1
]}

Loizou, PC ^{[1
]}

机构：

[1] Univ Texas, Dept Elect Engn, Richardson, TX 75083 USA

来源：

IEEE TRANSACTIONS ON SPEECH AND AUDIO PROCESSING | 2003年 / 11卷 / 05期

基金：

美国国家卫生研究院;

关键词：

multitaper power spectrum estimation; multiwindow covariance matrix estimation; perceptual weighting; spectral subtraction method; speech enhancement; subspace method;

D O I：

10.1109/TSA.2003.815936

中图分类号：

O42 [声学];

学科分类号：

070206 ; 082403 ;

摘要：

A new perceptually motivated approach is proposed in this paper for enhancement of speech corrupted by colored noise. The proposed approach takes into account the frequency masking properties of the human auditory system and reduces the perceptual effect of the residual noise. This new perceptual method is incorporated into a frequency-domain speech enhancement method and a subspace-based speech enhancement method. A better power spectrum/autocorrelation function estimator was also developed to improve the performance of the proposed algorithms. Objective measures and informal listening tests demonstrated significant improvements over other methods when tested with TIMIT sentences corrupted by various types of colored noise.

引用

页码：457 / 465

页数：9