For fixed sounds, panels learned from other people lowered mean prediction error in 20 iEEG patients and 16 fMRI adults

We tested whether responses from other people can choose useful calibration sounds. Separate public studies supplied signed iEEG LFP means and BOLD fMRI. Five patients and four MRI adults trained separate panels. A local code and panel lock preceded tests in 20 other patients and 16 other adults. These are new endpoints in reused public studies, not newly recruited cohorts. Each 32-sound learned panel had the same category counts as its acoustic reference. All future-person fits used the same ten-component acoustic regression. Mean sound-contrast error gains over acoustic selection were 2.161% and 5.856%, with 95% person Student intervals [0.919,3.402]% and [5.447,6.265]%. Gains over matched random panels were 2.200% and 8.010%. Mean skill over a constant forecast was 5.492% and 17.890%. The single shared gain-and-skill rule passed, p=0.000866092. Eight adults from another MRI study gave acoustic-comparison gain 3.935% with the unchanged MRI panels. The MRI studies share sounds. Common-sound and person reweighting kept positive ranges with fitted panels fixed. All negative people and descriptive controls remain. Reference target responses are required, so this is a known-library method for other people. It does not establish transfer to unknown libraries, prospective recording-time savings, a common neural source or causality. Clinical sampling, source screens, cohort differences, unequal features and measurement filters limit scope. The package preserves code, permitted inputs, source pins and terms, learned panels, residual summaries, all results, figures and repeat instructions. Original MRI arrays and audio stay external.

auditory cortex · stimulus selection · intracranial EEG · fMRI · cross-person prediction · optimal design