Speech reference gains were positive in 6 iEEG patients and 16 of 17 fMRI listeners; the mean rule failed

We tested whether one signal from other listeners adds later speech response prediction beyond sound and word timing. Public studies supplied separate human iEEG and fMRI cohorts with different stories. Three patients and eight listeners supplied fixed references. Six other patients and 17 other listeners supplied reserved tests. Gains were positive in all six patients and 16 of 17 listeners. Mean added test R2 was +0.008615 in iEEG (95% person t interval -0.000840 to +0.018070) and +0.052791 in fMRI (+0.034724 to +0.070857). The component sign tests passed. Each aligned group mean exceeded all 199 phase controls, which preserved reference spectra. The full fixed rule failed because the iEEG mean interval included zero. Positive bootstrap intervals do not replace that gate. The result supports a narrower common direction under this prediction procedure. It does not confirm the planned shared positive population mean. The reference can carry omitted acoustics, language, attention, or artifacts. No common neural cause or direct signal transfer is established. The benchmark uses separate reference and target people, retains negative scores, and accounts for person dependence. Clinical sampling, broad BOLD regions, full-record processing, fixed reference groups, and two fixed stories limit scope. The package contains compact inputs, source pins and terms, code, plans, all scores, figures, and full replay instructions.

human neuroscience · iEEG · fMRI · speech · intersubject correlation · prediction · independent subjects · negative result · reproducibility