Local sound level ratios did not pass a shared prediction test in six iEEG patients and 25 fMRI listeners

We tested whether current sound level relative to recent level in each frequency band improves later response prediction. Separate public studies supplied human iEEG during Podcast speech and fMRI during a Narratives story. Three patients selected a one-second context. The fixed test used six other patients and 25 fMRI listeners. Both models included absolute sound features, additive local history and word timing. Each model selected its penalty using internal validation. Mean added-minus-baseline test R2 was 0.0009034 in iEEG and 0.0145780 in fMRI. Four of six patients and all 25 listeners improved. The shared sign p was 0.34375, so the fixed shared rule failed. The iEEG bootstrap lower bound was near zero and its Student t interval crossed zero. A static compression control had higher mean scores in iEEG and lower mean scores in fMRI. This comparison is descriptive. An artificial static response also gained from ratio features. These results do not identify neural gain control. The contribution is a fixed negative cross-modal test and a reusable local sound context benchmark. Clinical selection, different cohorts and stories, regional BOLD means, full-run nuisance processing and the small electrical sample limit scope. The package preserves numerical inputs, source versions and terms, code, all failures and controls, figures, checks and repeat instructions.

iEEG · fMRI · auditory context · prediction · normalization · negative result · human neuroscience