Tests on excluded sound classes did not confirm shared gains from four spectral terms in 16 iEEG patients and 16 fMRI adults

We tested whether four mean spectral coefficients improve response predictions for an excluded sound class. Separate human studies supplied signed intracranial field means and BOLD fMRI. Each model fit within-class acoustic slopes in four classes and tested centered responses in the fifth. Seven base terms described pitch, loudness and harmonicity. A local lock preceded new endpoint tests in 16 iEEG patients and 16 MRI adults. Mean added gains were +0.339 and +1.054 points of centered response error. Pointwise 95% person Student limits were [-0.303,+0.980] and [+0.412,+1.696]. The fixed shared gain-and-skill test failed, p=0.14657593. Eight adults from another MRI study had an uncertain added gain. All negative people and classes remain. A crossed fixed-model sound/person range also included zero for primary MRI gain. Complete-refit gain ranges included zero in every study. We supply 199 complete nested refits per study, all draw weights and all person results. A fixed-site proof shows invariance to class-specific response offsets and nonzero site gains. Independent least squares, exact source-chunk checks and clean score replays passed. These are new endpoint tests in reused public data, not new prospective cohorts. Clinical sampling, different sounds, tasks, source screens, extraction builds and measurement filters limit scope. No positive cross-modal neural effect, absence of spectral coding or causal relation was established. The package gives a checked negative benchmark, efficient exact refit tools, permitted inputs, provenance, all scalar outcomes, figures and repeat steps.

human neuroscience · intracranial EEG · fMRI · auditory cortex · spectral shape · sound classes · nested validation · negative result