Abstract: This paper presents a microphone array speech enhancement algorithm based on statistical models. In order to estimate the magnitudesquared spectrum (MSS) of the first channel speech signal, it was assumed that the real part and imaginary part of the first channel speech signal spectrum were independent of each other and obeyed the Gaussian distribution with the same variance. The posterior probability of the MSS of the first channel speech signal was calculated by using the Bayesian formula, and the minimum mean square error (MMSE) estimation of the MSS of the first channel speech signal was obtained by using the mathematical expectation formula, and finally the spectrum angle of the first channel speech signal was estimated by using the spectrum angle of the first channel observed signal. The experimental results show that the proposed algorithm is superior to the three traditional speech enhancement algorithms.