Feature integration and auditory scene analysis (at)


Subject: Feature integration and auditory scene analysis
From:    at <WSWBUENGA.BU.EDU>
Date:    Mon, 24 May 1993 19:51:09 -0400

Dear Members, At the recent ASA meeting in Ottawa, Richard Pastore brought up Treisman's feature integration theory in a discussion of higher level auditory processing (2pMU3). As I (naively and grossly) understand its application to vision, this theory assumes that, in an analysis of incoming stimuli, several different, rapid processes (such as the determination of the color in each area of the visual field) are performed in parallel, and that at a higher level the output of these different processes (features) are integrated together to form the percepts of objects. This dichotomy is revealed in tasks such as determining whether a green figure is present in a background of non-green figures (rapid response time for criterion performance), and finding a particular green figure (say, a green X) amongst other multi-colored figures (for criterion performance, this occurs more slowly). The former is associated with the rapid processes (such as identifying the existence of a color), and the latter is associated with the slower, presumably "higher-level" process of forming conjunctions between features (such as color and shape). Thus, the name "feature integration". I believe several glimpses of feature-integration-like effects were apparent in talks at the Ottawa conference, in addition to those discussed by Dr. Pastore. In a talk by Bill Yost (3aMU1), it was noted that detecting the presence of modulation at one carrier was made difficult when other potential carriers existed in the stimulus. This is similar, as Dr. Yost pointed out, to results reported by Hall and Grose [JASA 90(6)] in which listeners (3 out of 4, at least) could identify the PRESENCE of modulation at modulation depth thresholds lower than those for which they could identify the carrier frequency of the modulation. Later in the same session (3aMU3), Bill Hartmann discussed difficulties with detecting a given mistuned harmonic of a complex when a second mistuned harmonic was present. These difficulties (associating modulation with a carrier, associating mistuning with a particular reference frequency) have the flavor of the feature integration tasks. That is, detecting whether modulation (or mistuning, in the latter example) exists is analogous to detecting the presence of a particular color (relatively low threshold), and detecting whether a particular carrier frequency is modulated (or mistuned) is analogous to detecting whether a form with a certain color is present (relatively high threshold). In one case, a particular feature must be detected. In the other case, one feature (carrier freq) must be associated or connected to another feature (modulation or mistuning). I do not mean to imply that Treisman's integration theory "explains" these results. Rather, to re-iterate, perhaps, some of Dr. Pastore's intent, I mean only to imply that the analogy between current work on auditory scene analysis and past work on visual scene analysis may well go beyond the Gestalt principles, that several examples of further analogy already exist, and that understanding the successes and failures encountered by theories such as Treisman's, before a great deal of these (successes, failures, and theories) are re-discovered or re-invented by we auditory scene analysis researchers, may lead to substantial gains for comparatively less effort (and that I can write a whopper of a run-on sentence). I hope this has been informative. -Bill Woods


This message came from the mail archive
http://www.auditory.org/postings/1993/
maintained by:
DAn Ellis <dpwe@ee.columbia.edu>
Electrical Engineering Dept., Columbia University