For two fixed sound panels, rank readers ordered speech above animal sounds and animal sounds above controls on average in iEEG and fMRI

We tested whether readers trained on human voices without speech place animal sounds between speech and other sounds. Separate public human studies supplied signed intracranial EEG and BOLD fMRI responses. Development used four people per modality. A fixed test used 20 other patients and 16 other adults. The primary input kept only response order across auditory sites. All speech, animals and about half of each source control class stayed outside fitting and tuning. Mean animal-versus-control AUC was 0.57609 in iEEG and 0.92272 in fMRI. Their 95% person t intervals were 0.54048–0.61170 and 0.89245–0.95298. Mean speech-versus-animal AUC was 0.73640 and 0.92243. The shared test required mean, person-sign and common test-label evidence for both boundaries in each modality. It passed correction for two planned claims, Holm p=0.021. The alternative order failed. Both observed boundaries were positive in 13/20 patients and 15/16 adults. Swapped control halves and five-site panel means retained positive estimates. An early electrical window, small early-area subsets and an electrical music-only control were uncertain. A crossed person-and-test-clip range for the electrical lower boundary was 0.50071–0.65131. Acoustic readers gave mixed results. A one-scalar model also gave the full order, so rank success does not count neural sources. These are new endpoint tests in reused cohorts and fixed panels. Clinical coverage, unequal clips, source site selection and measurement filters limit scope. No abstract voice code, common neural cause or causal effect is established. The package gives code, permitted inputs, exact source versions and terms, all outcomes, figures, audits and repeat instructions.

human neuroscience · intracranial EEG · fMRI · auditory cortex · voice perception · animal sounds · rank features · cross-modal generalization · independent controls