For 165 fixed sounds, category planes kept more mean within-group repeat covariance than shuffled planes in SEEG and fMRI

We tested whether directions that separate sound groups also keep stable measured differences within those groups. Separate human studies supplied SEEG high-gamma amplitude and BOLD fMRI. Each source had 165 sounds in three broad voice groups. Two patients and four MRI adults supplied development. A local lock preceded endpoint tests in six other patients and 16 other adults. Each fit used other sounds and one repeat to learn a rank-two category plane. Two other repeats tested within-group covariance. The comparison used 999 planes fitted to shuffled training labels. Mean excess covariance was +3.127 points in SEEG, with 95% person Student limits [+1.004,+5.251]. It was +2.646 points in primary fMRI [+2.172,+3.120]. One point is 0.01 of balanced training energy. One shared intersection test passed, p=0.03125. One negative SEEG patient remains. Eight added MRI adults gave +1.527 points [+0.956,+2.097]. The planes kept about one third of mean within-group repeat energy in both primary cohorts. Sound and person refits gave a SEEG range across zero. Short-baseline subtraction also left no clear SEEG effect. Pattern-length removal and specified time-order controls kept positive means. These are descriptive checks. The finding concerns fixed panels and source-selected processed records, not all new sounds, pure semantic coding or a common neural cause. Closely related SCI work was found after the first primary scores and is cited. This is a new controlled endpoint test in reused data, not new prospective confirmation. The package gives source pins and terms, code, complete results, figures, audits and repeat steps.

human auditory cortex · SEEG · fMRI · repeat covariance · voice categories · cross-modal tests · negative controls