Robust speech enhancement method based on fast Kalman filtering

The invention relates to a robust speech enhancement method based on fast Kalman filtering, which can fast solve a speech enhancement problem under an additive noise environment. The method comprises the following steps of: (1) performing frame separation on an input speech signal, and calculating the variance of the first frame as a noise estimated value; (2) entering loop iteration, operating a core iteration equation set of Kalman filtering by a system, and abandoning the calculation of an AR (auto-regressive model) coefficient and the calculation of a matrix, wherein the equitation set only needs to continuously update a first value of a state vector; and (3) outputting the speech signal after enhancement after the end of iteration by the system. According to the method disclosed by the invention, the operation is simple and convenient, the method is easy to implement by a computer, and the complexity of the system is greatly reduced and the robustness of the system is improved on the premise of not damaging the speech signal and not sacrificing the filtering effect.