A fixed test did not confirm shared music decoding after a voice projection in 19 iEEG patients and 16 fMRI adults

We tested whether music decoding remains after removal of response directions learned from human voices. Separate public studies supplied signed intracranial EEG field potentials and fMRI BOLD estimates. Four people per study supplied development. Fixed tests used 19 other patients and 16 other adults. The projection removed at least 90% of normalized training voice energy relative to a background centroid. Mean held-out music AUC was 0.55355 in iEEG, with a 95% person interval of 0.49078 to 0.61633. It was 0.98580 in fMRI, with interval 0.98011 to 0.99148. Shared music retention and relative-sparing claims both failed. Both Holm p values were 0.09125. A separate eight-person fMRI study gave mean music AUC 0.98218 after projection. It did not resolve electrical uncertainty. A post-primary energy reader retained some voice discrimination, but a three-claim correction failed and electrical sound uncertainty included chance. Exact algebra and noisy one-scalar constructions show why weak centroid reading cannot establish class erasure or independent neural sources. All 110,682 electrical means matched a fresh source rebuild. Source selection, unequal clinical coverage, distinct sound panels and measurement filters limit scope. No positive neural generalization or cause is claimed. The package preserves code, permitted inputs, source pins and terms, plans, all outcomes, figures and repeat steps.

human neuroscience · intracranial EEG · fMRI · music · voice · projection · negative result · reproducibility