Blind detection of exclusive source activity periods in reverberant acoustic environments

Abstract

Blind separation and dereverberation of acoustic sources is still considered a very challenging task despite many years of research and the availability of increasingly powerful computation engines. The complexity of the task can be significantly reduced for sources that exhibit sufficiently long exclusive activity periods (EAPs). EAPs are time intervals during which only one source is active and all other sources are inactive (i.e. zero). During EAPs the estimation of the underlying system parameters simplifies from a MIMO type to a SIMO type. The existence of EAPs is not guaranteed for arbitrary signal classes. EAPs occur very frequently, however, in recordings of conversational speech. In this paper we propose a new low complexity method for EAP detection which significantly outperforms earlier approaches.