Abstract: An improved speech enhancement algorithm is proposed for suppressing the background group noise in noisy speech. Compared with the traditional noise suppressor, the proposed method uses the soft-decision modification method to modify the log-spectral amplitude of noisy speech under the condition of uncertainty in speech signal presence. In this paper, an improved estimator for the priori signal-to-noise ratio and an effective estimator for the priori probability of speech absence are proposed, by which the conditional probability of speech presence can be obtained. Simulation results confirm the superiority of this method in noise suppression and speech enhancement under the background of Gaussian white noise and pink noise.