At 16 selected sites, a fixed repeat covariance rule failed a shared sound identification gain test in six SEEG patients and 16 fMRI adults

We tested whether repeat covariance learned from some sounds improves identification of other sounds beyond separate site scales. Public human SEEG high-gamma activity and BOLD fMRI came from separate studies. Each reader used 16 selected sites, spatial mean removal, separate fitting and test sounds, and a fixed half-shrink covariance. A local code lock preceded scores in six reserved patients and 16 reserved MRI adults. Mean reciprocal-rank gains were +0.001400 [-0.003979, +0.006779] and +0.010423 [+0.004754, +0.016092]. Brackets give 95% person Student limits. Gains were positive in 3/6 and 15/16 people. The fixed shared mean and sign test failed, p=0.65625. Eight adults from another MRI study gave mean gain 0.008168. The source reports distinct people across MRI studies. All studies reuse the sound library. A 24-site check had positive mean limits in both modalities; it cannot replace the failed main test. Short baseline subtraction gave a negative SEEG mean gain. An exact identity and an independent-error model check show how spatial mean removal changes the diagonal comparison. A fixed repeat-gain model gave a positive gain without additive repeat noise. These results do not establish a common neural cause or a general noise-normalization benefit. The contribution is a controlled negative benchmark, a comparison tool, source pins and terms, every outcome, figures and repeat instructions. Original neural arrays remain external.

SEEG · fMRI · human auditory cortex · noise normalization · covariance · sound identification · negative result · measurement