For 122 fixed sounds, mean matching within classes was above chance between one ECoG panel and two fMRI groups

We tested direct sound matching between a selected human ECoG panel and two separate fMRI studies. Twenty anchor sounds set site scales and distance profiles. The fixed test used 122 other sounds from 12 classes. Every false choice had the correct sound class. A local code lock preceded scores in 16 reserved B2021 adults. Mean rank AUCs were 0.60572 and 0.64096 in the two query directions. Chance is 0.5. The fixed test passed, two-claim Holm p=0.0002. An unchanged check in eight NH2015 adults gave 0.61199 and 0.63235, Holm p=0.0078125. A second acoustic-subtraction endpoint also passed. However, an exact category-only example rose from AUC 0.5 before subtraction to 1 after subtraction. Fitted acoustic subtraction can create fine matching and does not identify non-acoustic neural information. The original endpoint has an exact category-only guard. The electrical evidence is conditional on 32 distinct example profiles without released patient labels. There is no electrical patient holdout or population interval. Both MRI groups share the sounds and electrical reference. Small classes made 2774 of 4999 full target bootstrap draws undefined; defined-draw ranges are conditional. All negative person and class results remain. The package supplies source pins and terms, code, plans, complete scores and references, figures, exact limits and repeat instructions. This is measured correspondence for fixed sounds, not a shared neural cause or a semantic code.

human neuroscience · ECoG · fMRI · natural sounds · sound matching · representational similarity · acoustic controls · reproducibility