This paper deals with method of speech enhancement based on Karhunen-Loeve (K-L) and Wiener critical band spectral subtraction algorithm. The K-L algorithm is used twice to, first, track a spectral envelope of the contaminated signal and subtract sub-words - spectrally- homogeneous segments, and ,secondly, to construct an estimate of the spectral envelope of extracted sub-words using Wiener’s filter.

The paper is organized as follows: In section 2, we introduce the sub-word segmentation algorithm based on decomposition of covariance matrix in K-L transformation. Then in section 3, we describe our approach to speech parametr estimation in noise based on K-L transformation critical band optimal Wiener filter. In section 4, we compare our technique with the existing optimal K-L technique in adverse conditions, using both subjective and objective speech assessment measures. Result indicate that the proposed method provides better speech enhancement than the previous proposed algorithm.