Abstract: This study refined speech phase estimation to boost the performance of speech enhancement algorithms in low SNR environment. Firstly, the noise spectral magnitude estimator was derived from the speech magnitude estimator for the symmetry of the statistical model. then it was used to improve the phase estimation in each frequency bin, after proving the importance of phase estimation. Finally, a modified estimator for the a priori SNR is proposed. The simulated experiments indicate that significant improvement can be achieved. For the speech enhancement in low SNR environment, the algorithm obtains an obvious improvement in speech quality, with little musical noise retained.