An alternate binary mask is constructed to improve the speech intelligibility and quality based on constraints of the magnitude spectrum. Motivated by previous studies of speech intelligibility obtained using processing strategies based on ideal binary masks, a new method for deriving a mask is proposed that separates noisy speech into time-frequency channels. The binary mask decisions are made on the basis of speech constraints imposed by the spectral subtraction Gain-induced distortions. All time-frequency channels satisfying the constraints are retained while time-frequency channels violating the constraints are discarded. The speech signals degraded at various signal-to-noise levels (-5dB, 0dB, +5dB) using babble and street makers are processed by the proposed mask and are presented to the normal hearing listeners in experiments measuring speech intelligibility. The results revealed significant improvements in speech intelligibility and quality even at low SNR levels.