At selected auditory sites, a voice contrast kept repeatable nonvocal sound differences in six SEEG patients and 24 fMRI adults

We tested if a direction fitted to separate vocal from nonvocal sounds also keeps repeatable scores within the source nonvocal group. Separate public human studies supplied SEEG high-gamma activity and BOLD fMRI. Development used two patients and four adults. A local lock preceded tests in six other patients, 16 B2021 adults and eight NH2015 adults. Five sound folds excluded every test sound from direction fitting. Repeat 1 fitted the direction; repeats 2 and 3 tested it. Spatial means were removed. Mean nonvocal rank scores were 0.6480, 0.7058 and 0.5561. Their 95% person intervals were [0.4040,0.8919], [0.6495,0.7621] and [0.4590,0.6532]. Every person also had positive scaled cross-repeat covariance and class AUC above 0.5. One shared intersection test passed, p=0.015625. An exact equal-mean error example gave expected rank 1/4 and covariance zero. It prompted the required covariance gate before evaluation. Unit-length and 16-site checks retained positive means. A specified short-baseline subtraction left the SEEG rank interval across zero. A longer baseline was unavailable. The random-label benchmark gave no clear SEEG advantage for the voice direction. These limits remain. The result concerns a measured class-direction score in selected records. It does not identify perceived voice likeness, latent source count, a common mechanism or causality. Clinical sampling, separate cohorts, source screens using repeats, fixed sounds and unequal filters limit scope. The package preserves code, source pins and terms, all scores, uncertainty, failures, figures and repeat steps. Original neural arrays remain external.

human auditory cortex · SEEG · fMRI · voice selectivity · repeatability · independent repeats · class direction · source selection