Dimensions of auditory segregation : What do they tell us about levels of auditory processing ?

Perceptual segregation of multiple, simultaneous auditory streams has been becoming an area of interest to several investigators since the publication of Bregman’s (1991) landmark book on the subject. The principal motivation behind these investigations has been to find a closure on the classic, and invariably elusive, problem of the auditory segregation of simultaneous speech signals, i.e., the “cocktail-party effect.” Search for psychophysical underpinnings of this unique human ability, however, has been by-and-large limited to research on the segregation of concurrent harmonic complexes, especially vowels (Marin and McAdams 1991; Carlyon 1992; de Cheveigné 1993; Assmann 1996). The present work attempts to approach the problem from a more comprehensive point of view and take a look at how the various auditory dimensions contribute to, and interact in, the segregation of concurrent sounds. Earlier (Divenyi 1995), we proposed a framework for auditory segregation of sounds concurrently emanating from several sources, that incorporated three main, “cardinal” dimensions: the acoustic characteristics of the signal emanating from each source (including short-term envelope fluctuations down to modulation frequencies of about 20 Hz), the temporal pattern generated by syllabicand subsyllabic-rate envelope fluctuations within each source, and the spatial location of the sources. For a necessary and sufficient criterion of segregation, we proposed to adopt the dual requirement of (1) having to resolve the difference between the concurrent sources along the diverse dimensions, and (2) correctly associating the particular values perceived along the various dimensions. Thus, if Talker A says X and Talker B simultaneously says Y, even if A, B, X, and Y are correctly identified, the report that “Talker A said Y and Talker B said X” indicates an incorrect association and will not be accepted as signaling segregation. In addition, we postulated that, at some segregation threshold, differences between concurrent signals should trade off against one another.