Poster C111, Sunday, March 26, 5:00 – 7:00 pm, Pacific Concourse
Cortical Networks for Intelligible Speech Identified with Reverse Correlation
Jonathan Venezia1, Gregory Hickok2, Virginia Richards2; 1VA Loma Linda Healthcare System, 2University of California, Irvine
Cortical networks for intelligible speech have been described largely in terms of region-level preferential responses to speech versus non-speech sounds. Here, we present an auditory fMRI paradigm that quantifies the degree to which voxel-wise “speech receptive fields” reflect the acoustic information most important for intelligibility. Our procedure works by filtering spectrotemporal modulations (STMs) from the speech signal at random over many trials and relating the filter patterns to an outcome measure using reverse correlation. In the present fMRI experiment, 10 listeners were presented with STM-filtered sentences and asked to make subjective (yes-no) intelligibility judgments. Behavioral response rate was held constant (50% “yes”) using adaptive methods. Voxel-wise single trial response amplitudes were estimated and reverse correlation was used to determine: (a) the STMs that reliably produced “yes” responses (perceptual receptive fields); and (b) the STMs that reliably produced stronger BOLD responses (neural receptive fields). Neural receptive fields were derived independently of perceived intelligibility and were based only on which STMs produced the highest-amplitude single-trial BOLD responses. We then used Mutual Information (MI) to quantify the statistical similarity between neural receptive fields and intelligibility-based perceptual receptive fields. Significant responses were identified using permutation testing (P < 0.05). The largest MI responses were observed bilaterally in the transverse temporal sulcus, and in distinct clusters of the posterior, middle, and anterior superior temporal gyrus/sulcus. Weaker but still significant MI responses were observed in the left inferior frontal lobe.
Topic Area: PERCEPTION & ACTION: Audition