In this paper, we first define the scenario of a generic acoustic human/machine interface and then formulate the according fundamental signal processing problems. For signal reproduction, the requirements for ideal solutions are stated and some examples for the state of the technology are briefly reviewed. For signal acquisition, the fundamental problems ask for acoustic echo cancellation, desired source extraction, and source localization. After illustrating to which extent acoustic echo cancellation is already a solved problem, we present recent results for separation, dereverberation and localization of multiple source signals. As an underlying motivation for this synoptic treatment, we demonstrate that the considered subproblems (except localization) can be directly interpreted as signal separation or system identification problems with varying degrees of difficulty, which in turn determines the effectiveness of the known solutions.

Download