Towards a brain-machine system for auditory scene analysis

Abstract

In hearing aids, noise suppression algorithms that rely on spatial cues tend to improve the intelligibility of speech in noisy environments [1], [2], [3], [4]. Unfortunately, the location of target and noise sources can change rapidly in a natural, everyday acoustic environment. In fact, depending on what the listener is attending to, one source may be considered noise in one instant, and then considered the target in another instant. Adaptive filtering attempts to track the target source, but it is successful only under a set of simplifying constraints [5], [6]. It is much more effective to allow the user to determine the direction from which the target sound is coming.