A Hybrid Adaptive Beam-former and Post-filter for Speech Enhancement in Real Environments

Diversity of the noise fields in real environments makes difficult to design an efficient de-noising system. This paper addresses the problem of speech enhancement in real environments. Towards this goal, we present a hybrid scheme based on the adaptive beam-forming and post filtering. In order to capture the adaptive conditions of the environment, we adopt two algorithms: steepest descent and least mean square. We also consider the application of the post-filters on the output of the beam-former. As an advantage, the proposed structure not only produces some controlled nulls to reduce the directional noises, but also adaptively minimizes the power of diffuse and uncorrelated noises. The study of the convergence of the proposed algorithms and also the behavior of the diffuse noise in real conditions are other issues considered in this paper. Numerical results of our comprehensive implementation confirm the efficiency of the proposed scheme in suppression of the real environmental noises.