Presentation

Sahil Luthra¹, Adam Tierney², Lori Holt¹, Frederic Dick^2,3; ¹Carnegie Mellon University, ²Birkeck, University of London, ³University College London

When listening to speech, listeners must selectively attend to certain sources (e.g., a particular talker) and also focus on informative acoustic dimensions within a source (e.g., particular frequencies that may be informative for determining what is being said). A common but untested assumption is that source-based selective attention and dimension-based selective attention are supported by common mechanisms. In this ongoing fMRI project, we directly compare source-based and dimension-based auditory selective attention across a naturalistic ‘cocktail party’ listening task in which two simultaneous audiobook recordings (one read by a male talker, one by a female talker) of Alice’s Adventures in Wonderland are presented to place demands on spatial- (source) versus frequency- (dimension) selective auditory attention. All listeners (target N=25) are familiarized with the story prior to an fMRI scan by listening to clear, single-talker recordings of the to-be-attended portion of the story. In half of the fMRI runs, listeners hear speech stimuli spatialized via interaural level differences, such that there is a 6dB RMS left > right ear difference for one talker, with a corresponding right > left ear difference for the other talker. In the other half of runs, talkers are instead spectrally segregated; one talker’s voice is band-pass filtered to higher frequencies (1800 to 8000 Hz), and the other’s to low frequencies (300 to 1600 Hz). Both spatial and spectral signal processing manipulations result in intelligible speech from both talkers, although listening is challenged by the competing talker. Listeners are instructed to attend to a particular talker (male, female) over the course of each fMRI run and, every 15-30 seconds, the talkers switch locations or spectral bands. To behaviorally assess a listener’s ability to selectively attend to a target talker, listeners respond to occasional semantic anomalies in the story (e.g., “And what good is a book without pictures and pink eyes?”). Participants also complete a tonotopy mapping task with non-speech stimuli, which will serve to define frequency-band-selective regions-of-interest. We hypothesize that a comparison of attending to the spectrally segregated talkers will reveal considerable fine cortical spatial segregation within “traditional” tonotopically mappable auditory regions, and also may unveil additional cortical regions where attention to different spectral bands is spatially organized, and are otherwise difficult to characterize with short natural sounds or more sustained artificial tone streams. This comparison also permits us to test the hypothesis that cognitively or linguistically mediated changes in neural activity can be spatially organized in cortex along a fundamental perceptual dimension, frequency. Conversely, and based on past literature, we expect these same regions to exhibit overlapping neural signals associated with spatially segregated talkers. However, we also hypothesize that the larger functional networks associated with these two types of sustained auditory attention may be spatially interdigitated. Overall, this project will clarify the neural mechanisms that support both source-based and dimension-based auditory selective attention in processing of naturalistic speech stimuli.

Topic Areas: Perception: Auditory, Perception: Speech Perception and Audiovisual Integration