A Two-Step Noise Estimation Algorithm for Noisy Speech Enhancement

Noise estimation is an important part for noisy speech enhancement due to its momentous effect on the intelligibility and quality of the enhanced speech. In this paper, an effective noise estimation algorithm is presented by combining the minimum statistics estimation and Gaussian model assumption. In contrast to other methods, the proposed approach works in two steps. The noise power estimated by minimum statistics method in the first step suffers some bias which is removed by the second step of the proposed approach using Bayes theorem. As the noise estimation is refined in the second step, more accurate estimation can be obtained. The performance of the proposed approach is evaluated by objective and subjective tests under various noise environments and found to yield better results compared with conventional MS -based estimate.