Passage mean removal changed mean sound amplitude tracking skill by minus 0.18 points versus ordinary fitting with matched validation in ten offline SEEG readers with eight pairs and 400 ms delay

We tested whether removing training passage means improved sound amplitude tracking in public SEEG data. The test used 95 fixed passages from ten listeners, eight bipolar pairs per person, 10 Hz inputs, and a 400 ms retrospective delay. The ordinary and centered training fits used the same centered validation score. Mean tracking skill was 11.34% for the ordinary fit and 11.16% after training mean removal. The paired change was minus 0.18 percentage points. The descriptive 95% person bootstrap interval was minus 0.78 to plus 0.31 points. The test did not establish a gain or equivalence. The older ordinary comparator retained 10.27% tracking skill after passage means were removed for scoring. Passage mean mismatch accounted for 11.39% of its absolute squared error. Quality and scale checks gave the same limited conclusion. Test means were used only for completed offline scores. Shared sounds and a reused small cohort limit inference. The package preserves code, derived inputs, source versions and range hashes, raw predictions, fitted models, person and passage results, three figures, and reproduction instructions. Fresh source checks reproduced one selected passage for each person. Independent solver checks and both full analysis repeats passed. This is a heard-speech readout study. It does not measure speech intention, live output, or brain stimulation.

SEEG · sound amplitude tracking · passage mean removal · brain-computer interface · ridge regression · offline neural readout