In this paper, a perceptual subspace speech enhancement method using masking property of the human auditory system and variance normalization is presented. The simultaneous masking property of the human auditory system is used while deciding the gain parameters of the algorithm. Spectral Domain Constrained estimator was employed in determining the filter coefficients, and colored noise was handled by replacing the noise variance by Rayleigh quotient. Normalized variance was used as control parameter which adaptively decided the amount of filtering to be performed. Adaptively varying the control parameter enabled the proposed algorithm to perform better in non-stationary colored noise environments, compared to some of the other existing algorithms. The waveforms and spectrograms indicate that the proposed algorithm removes most of the noise without distorting the speech signal. The results obtained in terms of objective parameters (wcep,WSS,SNRLOSS and SNRLESC) and informal listening test support the claim.