Speech enhancement using non-linear extended Kalman filter to train multi-layer perceptron with backpropagation

The development of algorithms that enhance speech quality is crucial because of their wide use in various applications where noise-free signals are needed. As a contribution in this field to tackle the challenging problem of improving the quality of noisy speech, we propose in this paper an approach that deploys an extended Kalman filter (EKF) to train a multi-layer perceptron (MLP). Since speech damaged by noise is usually available, this approach is particularly useful for solving this issue. Experiments employing the NOIZEUS database demonstrate that the method we talked about above yields superior subjective and objective findings when compared to other optimization roads.