Target Speech Enhancement in Presence of Jammer and Diffuse Background Noise

摘要

This paper presents a method for enhancing a target speech in the presence of a jammer and background diffuse noise. The method is based on frequency domain blind signal separation (FD-BSS). In particular, the permutation resolution is done using both the direction of arrival (DOA) information contained in the estimated filters and some statistical features computed on the estimated signals. This enables the separation of the target speech, the jammer and the diffuse background noise which is not possible if using only the DOA or the statistical features. Since in presence of diffuse noise, FD-BSS cannot provide a good estimate of the target speech a channel wise modified Wiener filter is proposed as post processing to further enhance the target speech.