Ross Maddox - rkmaddox@uw.edu
Willy Cheung - wllychng@uw.edu
Adrian KC Lee - akclee@uw.edu
Department of Speech and Hearing Sciences &
Institute for Learning and Brain Sciences (I-LABS)
1715 Columbia Road NE
Portage Bay Bldg, Rm 206
University of Washington, Box 357988
Seattle, WA 98195-7988
Popular Version of Paper 5aPP4
Presented Friday morning, May 18, 2012
163rd ASA Meeting, Hong Kong
Imagine you are in a crowded restaurant. Dishes are clanking, music is playing in the background, and your children are fighting over who gets to play with your new electronic gadget. Meanwhile, you notice two coworkers just sat down at a table near you. Through the cacophony, you hear your name uttered, grabbing your attention. You focus in on your coworkers' conversation (hoping that they are saying something nice behind your back).
Selectively paying attention to different conversations can be hard in the above scenario (commonly referred to by scientists as the "cocktail party effect"), but most normal-hearing listeners can do it quite successfully. So how many simultaneous sound streams in an environment does it take to break down our ability to selectively attend (or covertly eavesdrop on) just one conversation? Are there any practical applications once we answer this question?
Most studies measure how degrading the target sound affects how well subjects can perform the task; here, however, we tried to make the target as easy to distinguish from the other sounds as possible by giving it a distinct location, pitch, and timing. We then asked listeners to selectively listen to one stream of spoken letters amidst up to 11 other sound streams. Subjects performed very well, especially if we informed them (by playing them an example letter from each stream before the trial started) where, when, and to which pitch they should listen. Once they "locked on" to the stream of interest, they could direct their attention and perform the task (specifically, pressing a button to indicate that the letter to which they were listening changed to an R for one repetition).
Why do we want to push the limits of auditory attention like this? This study is one of a new series of experiments designed to discover how we can optimize sound presentation for an auditory-based Brain Computer Interface (BCI). BCI allows users, such as locked-in patients (who are fully conscious but cannot control their bodies), to control computers and machines by reading their brain signals and converting them into commands. Most BCI are currently visual or motor-imagery based (for example, the patient thinks about moving her left arm and the BCI recognizes a specific brain pattern over the right motor cortex). An auditory BCI would work by presenting a user with several sounds at once and decoding which one she is listening to, creating a sort of "menu" system. The more sounds present at one time, the more options the user has for BCI commands. The results of this study, then, bring us one step closer to an optimized auditory BCI design that is useful in a medical rehabilitation setting where other forms of BCI are not appropriate. Furthermore, this could also have interesting commercial applications.