Model matching for signal enhancement

In many advanced signal processing applications including acoustic signal enhancement, signals are not known a priori, except for some general statistical properties. These properties are typically encapsulated in statistical models. It is then intuitively expected that by matching these models, target signals can be recovered. Consequently, the aim of this paper is to propose a new model-matching-based signal enhancement approach, which employs the Kullback-Leibler divergence to design new signal enhancement filters. We particularly focus on the single-channel case where the desired and undesired signals have Laplacian and Gaussian distributions, respectively.