In one SEEG patient, high-frequency responses to repeated sentences declined mainly after onset; time-in-session remains a possible cause

We asked whether the 70–150 Hz high-frequency activity (HFA) response of stereo-EEG (SEEG) contacts declines when a spoken sentence is repeated. In one patient (OpenNeuro ds004993, sub-W2), we analyzed 10 TIMIT sentences presented 10 times each across 98 bipolar contact pairs. Thirteen pairs met the response criterion, all on two adjacent right temporal shafts. Across those pairs, the mean slope was −0.059 SD per repeat; 12 of 13 slopes were negative. Shuffling repeat order within each sentence gave a two-sided permutation p = 0.0004. The predicted change from repeat 1 to repeat 10 was −0.54 SD. The first presentation exceeded the mean of presentations 2–10 by 0.57 SD. The decline was present in the late window (0.3 s to sentence end; slope −0.066, 13/13 negative, p = 0.0005), but not in the onset window (0–0.3 s; slope −0.014, p = 0.30). Whole-sentence mean responses did not reliably distinguish the 10 sentences in this patient (split-half median r = 0.02), so this measure should not be used as a noise-ceiling estimate. The result is limited to one patient, and the responsive pairs are not independent. Repeat number is confounded with time in session; the planned control analysis in a second patient was not run because the EDF download did not complete. A fatigue or signal-drift effect is therefore not excluded. The findings indicate that including the first presentation may bias noise ceilings that assume equal expected responses across repeats; removing that presentation or modeling repeat number is advised. The archived package contains the analysis plan, NumPy-only code, results, sidecar files, source URLs, and hashes. The exact OpenNeuro dataset release used was not recorded.

stereo-EEG · high-frequency activity · repetition suppression · speech · noise ceiling · TIMIT · OpenNeuro · within-subject analysis