Sound model subtraction lowered mean coupling in nine iEEG patients and 44 new fMRI listeners; an fMRI check gave negative mean prediction scores

We reproduced an exploratory iEEG observation and tested it in 44 previously unused fMRI listeners with two other stories. A fixed sound and word timing model fit earlier data and predicted later test data. We subtracted only its feature component, then measured within-person correlation change. Mean Fisher-z coupling decreases were 0.03565 in nine previously used iEEG patients, 0.06186 in Black fMRI, and 0.08074 in Forgot fMRI. New-story 95% person intervals were 0.02877 to 0.09537 and 0.03101 to 0.13130. The independent fMRI intersection p was 0.00121. The same 44 people heard both stories; we did not count 88 records as independent people. All primary conditions passed. However, Black prediction quality remained uncertain. A planned fMRI penalty check also lowered coupling while mean test R-squared was negative in both stories. Thus, correlation decrease alone did not validate the subtraction as a correction. This study confirms transfer of the defined numerical effect to new fMRI data; clinical evidence remains prior-informed reproduction. It does not recover true connectivity or establish causality. The package gives complete compact inputs, fixed code, source hashes, model fits, person uncertainty, all negative results, figures, and repeat instructions.

human neuroscience · iEEG · fMRI · functional connectivity · independent confirmation · speech · prediction quality · reproducibility