Speech/song reading stayed above chance in the weakest fitting half of selected sites in six SEEG patients and 16 fMRI adults

We tested whether speech/song reading survives removal of half of selected auditory sites. Each fit removed sites with the largest absolute standardized class mean contrast. A centroid reader used the rest. It predicted other sounds in other repeats. Separate public studies supplied SEEG high-gamma responses and BOLD fMRI. Two patients and four MRI adults supplied development. A local plan and code lock preceded tests in six other patients and 16 other adults. Mean AUC and 95% person Student limits were SEEG 0.73609 [0.61321, 0.85896] and fMRI 0.91289 [0.89287, 0.93291]. The single shared test passed, p=0.00216297. The test required positive mean evidence and a common sound-label reference in each modality. Controls, negative people and weaker retention fractions remain. A third MRI acquisition had mean AUC 0.83488; overlap of its people with the primary MRI study was not verified. Selection uses noisy fitting preferences. The result does not identify truly nonselective sites or independent neural sources. A one-scalar model supports deletion reading. Clinical sampling, source site screens, unequal coverage, age, task, acoustics and vascular filters limit interpretation. The contribution is a fixed deletion benchmark in reused human cohorts. It does not show a common cause or a neural lesion effect. Code, source pins and terms, all scores, figures and repeat steps are supplied. Original neural arrays stay external.

human neuroscience · SEEG · fMRI · auditory cortex · speech · song · site selection · readout robustness