Acoustic models predicted response differences within omitted sound categories in one iEEG panel and eight fMRI adults

We tested whether acoustic features predict response differences within a sound category that was absent from fitting. Separate public human studies supplied one selected iEEG panel and fMRI from eight adults. All fits used 165 common sounds and 12 source labels. A fixed ridge model used six spectrum and 63 modulation features. Each fold excluded a whole category. Evaluation removed category means. BOLD fits used scans 1 and 2; test responses came from scan 3. Mean within-category correlation was 0.285121 in 32 distinct electrical profiles and 0.228828 across BOLD people. The 95% BOLD person t interval was 0.208762 to 0.248894. Both refitted sound-label tests gave Monte Carlo p=0.001; all eight BOLD people were positive. The fixed conjunction p was 0.00390625. Secondary refit sound bootstrap ranges were 0.208 to 0.347 and 0.182 to 0.251 with people fixed. BOLD site scores and one category-person score were negative. All results remain. Missing electrical patient labels limit that evidence to the fixed panel. This is a new endpoint in reused data, not new-cohort confirmation. The result supports acoustic prediction across the supplied labels in these selected measurements. It does not show a common neural code or causality. Clinical sampling, source selection, shared stimuli, broad labels and measurement differences limit scope. The package gives code, source versions and terms, plans, uncertainty checks, all derived scores, figures and repeat instructions. Original arrays remain external.

auditory cortex · intracranial EEG · fMRI · natural sounds · category generalization · encoding models · cross validation