Relations to other sound classes supported matching between 32 fixed iEEG profiles and fMRI in 28 adults

We tested whether response relations to other sound classes support fine sound matching. A fixed panel of 32 distinct source-selected intracranial profiles and two separate fMRI studies supplied responses to 165 clips. Targets came from ten classes with at least five clips. Other classes supplied all anchors. Class means were removed. Same-class candidate ranks were averaged over 158 targets. In 20 B2021 and eight NH2015 adults, mean AUC was 0.7112 and 0.6569, with 95% person limits [0.6985,0.7238] and [0.6372,0.6765]. All 28 scores were above chance. The shared rule passed, Holm p=0.00781250. A later fixed repeat test in eight identified SEEG patients passed, mean AUC 0.7159, 95% person limits [0.6870, 0.7448], Holm p=0.00781250. MRI repeat checks remained descriptive. The electrical reference lacks patient labels, so cross-study uncertainty conditions on that panel. The separate repeat test does not repair that population limit. Acoustic references had higher cohort mean ranks. An unfitted rank-one gain model also scored above chance. Exploratory removal of one or three anchor modes kept positive scores in all MRI people. Thus matching does not establish a semantic code, unique item codes, neural rank or a cause. The package preserves code, source pins and rights, local plans, all results, uncertainty, figures and repeat steps. Original neural arrays remain external.

human neuroscience · intracranial EEG · SEEG · fMRI · sound matching · representational similarity · public data · replication