A 0.1 s sound contrast term failed fixed prediction tests in six iEEG patients and 25 fMRI listeners

We tested whether recent sound contrast improves prediction of human responses to speech. Public inputs came from separate intracranial EEG (iEEG) and fMRI studies with different stories and people. Three iEEG patients selected a 0.1 s context window. A local plan then tested six other patients and all 25 fMRI listeners. The feature combined current log sound level with inverse recent log-level standard deviation. Models controlled for sound bands, static level, sound changes, word timing, recent mean level, and separate delay ranges. Mean added test R2 was -0.0003102 in iEEG, with a 95% person interval from -0.0006555 to -0.0000589. It was -0.0135849 in fMRI, with an interval from -0.0193828 to -0.0086379. All six patients had positive fitted coefficient sums but negative prediction changes. The fMRI coefficient direction was uncertain. The planned shared positive rule failed; its intersection p was 1. A later saved-fit audit separates test alignment from added prediction energy. An unplanned fMRI scan-average check also gave a negative mean gain. These results concern a fixed model and contrast proxy. They do not establish absent biological gain control or a shared causal process. Clinical sampling, regional BOLD pooling, offline processing, and two fixed stories limit scope. The package preserves compact inputs, source versions and terms, code, all outcomes, figures, and repeat instructions.

human neuroscience · intracranial EEG · fMRI · auditory cortex · contrast gain control · negative result · speech prediction · reproducibility