Abstract | ||
---|---|---|
A perceptually motivated subspace based approach is pro- posedforenhancementof speechcorruptedbycolorednoise. The proposed approach takes into account the frequency masking properties of the human auditory system and re- duces the perceptual effect of the residual noise. Objec- tive measures and informal listening tests demonstrated im- provements over other subspace-based methods when tested with TIMIT sentences corrupted with speech-shaped noise. the frequency spectrum of the residual noise. Motivated by the perceptual weighting techniques used in the analysis- by-synthesis speech coders (5), we propose a new subspace based speech enhancement approach, which takes into ac- count the frequency masking properties of the human audi- tory system. We also address the issue of obtaining a good estimate of the covariance matrix, an issue which is very important for the performance of subspace-based speech enhancement methods. McWhorter and Scharf's work (6) provided a new viewpoint for the problem of covariance estimation. They showed that the commonly used covariance estimators are special cases of multiwindow estimators and that a special type of window may be used to improve the estimator's per- formance. Our proposed approach integrated this multiwin- dow covariance matrix estimator. This paper is organized as follows. In section 2 the proposed perceptually-motivated subspace speech enhance- ment approach is presented. In section 3, implementation details are described. Experimental results are presented in section 4, and the conclusions are given in section 5. |
Year | Venue | Keywords |
---|---|---|
2002 | INTERSPEECH | covariance estimation,covariance matrix,frequency spectrum,analysis by synthesis |
Field | DocType | Citations |
Speech enhancement,Pattern recognition,Subspace topology,Computer science,Covariance intersection,Speech recognition,Artificial intelligence | Conference | 5 |
PageRank | References | Authors |
0.65 | 6 | 2 |
Name | Order | Citations | PageRank |
---|---|---|---|
Y. Hu | 1 | 1098 | 69.42 |
Philipos C. Loizou | 2 | 991 | 71.00 |