Auditory scene analysis: Theory and phenomena

Abstract

Many natural listening environments face the auditory system with a jumble of overlapping sound-producing events, from which it must recover separate descriptions of the individual sounds (auditory scene analysis). Its most basic method is ‘‘primitive’’ auditory grouping, a pre-attentive process that analyzes the incoming signal into components and then uses certain acoustic relations among them to link them into sets that later-acting recognition processes will prefer to treat as coming from separate events. Components (or features) will be linked when they exhibit relationships to one another that would be very improbable had they actually come from unrelated events. The evidence for grouping, derived from different relations, is allowed to compete and collaborate, in the linking of subsets of components or features. In a mixture, the listener tends to hear these subsets as distinct but overlapping sound sequences, each with its own properties. The perceived qualities that can be affected by this organizing of data include the melodic and rhythmic qualities of the perceived sounds, their pitches, timbres, numerosity, temporal order, and even their perceived spatial positions.