In this paper, we propose a fast time-frequency mask technique for blind source separation in order to separate a mixture of two input sounds in single signal automatically. Mostly previous methods utilize a linear sensor array, and therefore they cannot separate symmetrically positioned sources. To overcome such problems, we first define two features which are normalized level-ratio and phase-difference. Next, we use our method to decrease Direction of Arrival (DOA), this can reduce the variance of features so that it can reduces iterations of k-means. Finally, according to the clustered features, a mask is generated. Our method does not require any prior information or parameter estimation and we have made a real demonstration system. We use Signal to Distortion Ratio ( SDR ) and Signal to Interference Ratio ( SIR ) to compare our method. Then we present hardware design. Hardware design uses TSMC 90-nm CMOS process. As a cost-effective result, it consumes about 120K gates and executes with frequency of 10MHz. The power consumption is only 2.92 mW with low power design considerations.