Skip to Main Content
A noise autocorrelation estimator based on the optimal first-order smoothing recursion and minimum energy algorithm is described in this paper. The estimator can be combined with any speech enhancement algorithm based on subspace which requires an accurate estimate of noise autocorrelation. Unlike the other approaches used to estimation noise autocorrelation, the proposed approach estimated it from the autocorrelation of noisy speech directly. The simulation results show that the proposed estimator performed better than the traditional estimators, especially in the nonstationary noise environment.