The speech corrupted by additive noise can be enhanced by the Karhunen-Loeve transformation (KLT) method. However, musical noise is usually introduced by this method. This paper studies an improved perceptual KLT (IPKLT) method for speech processing, which is based on the combination of PKLT algorithm and Wiener filter with noise statistics being estimated by minimum tracking method. The Wiener filter is formed using the signal-noise-ratio (SNR) formula in subspace domain. Then the eigenvalues of the clean speech covariance are obtained through it. Simulation results show that the SNR gained with the proposed algorithm is higher than that obtained using conventional KLT and PKLT methods for the speech contaminated by babble and train noise. Moreover, the musical noise is suppressed effectively.