Ranks matched nine vocal clips in selected SEEG and fMRI data; full-panel selection gave 62.9% matching in a noise model

We tested whether the order of auditory site responses can match non-speech vocal clips across repeats. Separate public studies supplied SEEG high-gamma activity and two fMRI cohorts. Only 128 source nonvoice clips set site means and scales. A rank reader kept site labels and removed exact response magnitudes. Nine other human vocal clips supplied the test. Development used two patients and four MRI adults. A local code lock preceded tests in six other patients, 16 B2021 adults and eight NH2015 adults. Mean rank matching was 65.35%, 62.66% and 62.09%, with 95% person intervals [60.76,69.93]%, [60.79,64.53]% and [58.89,65.29]%. Chance was 50%. All 30 scores were above chance. The fixed shared rule passed, Holm p=0.03125. A fixed claim of 90% excess-score retention against a continuous reader failed. Original source screens used the full sound panel. A model added after human scoring tested this limit. In 300 records with only independent Gaussian errors, selecting 16 of 2,048 sites for full-panel repeat agreement gave 62.86% rank matching. Selection with calibration clips alone gave 49.83%. This model was not fitted to human screens or scores. It does not estimate their bias or explain their cause. A second exact example gave perfect rank matching from one scalar input. Thus, the human evidence supports a reader for selected processed measurements. It does not establish signal specificity, a shared neural code, a source count or causality. Separate cohorts, clinical sampling, source selection, common clips and unequal sensing limit scope. The package gives source pins and terms, code, all results, plans, figures, numerical audits and repeat steps. Original neural arrays stay external.

SEEG · fMRI · sound identification · rank correlation · source selection · reproducibility