For fixed sounds, greater speech and song pattern spread repeated in two fMRI groups; a test in six SEEG patients failed

We tested whether speech and song have more repeatable pattern spread than other human voices. Separate public studies supplied human SEEG high-gamma responses and fMRI BOLD responses. Each person supplied three repeats or response estimates, nine other-voice clips and 28 speech or song clips. One repeat set site scales from 128 nonvocal sounds. The other two set distances after spatial means were removed. Development used two patients and four adults. A local code lock preceded tests in six other patients and 16 other adults. Mean spread differences were -0.00481 and +0.16879. Their 95% person Student limits were [-0.26155,0.25192] and [0.12722,0.21036]. The shared test failed, p=0.53125. Eight adults from a separate fMRI study gave +0.07317 [0.02512,0.12121]. That study adds people but uses the same sounds. Its person-and-clip range included zero. A 24-site rule reversed its mean. Negative people and all planned checks remain. An exact one-scalar model gives high direction agreement with either more or less spread. Named audio features had lower total speech/song variance but allowed a linear projection with higher variance. Thus neither direction agreement nor total acoustic variance identifies the neural explanation. The contribution is a checked group-variation assay, a failed shared test, and compact derived pair data. Clinical coverage, source selection, few clips, different cohorts and sensing filters limit scope. No positive neural generalization across modalities or causal effect was confirmed.

human SEEG · fMRI · auditory cortex · natural sounds · crossvalidated distance · repeatable pattern spread · negative result · sound sampling