In one SEEG patient, sentence repetition and session time were not separable predictors of speech HFA

A prior analysis of OpenNeuro ds004993 found that the 70–150 Hz high-frequency activity (HFA) response to spoken TIMIT sentences decreased across repeats in sub-W2, but could not separate repetition from time in session. We used the randomized, interleaved order of 10 sentences, each presented 10 times, to test both effects, and used sub-W3, who heard 126 different sentences once each, as a time-only control. In sub-W2, within-sentence log repeat and time correlated at r = 0.94; only 11% of log-repeat variance was unique. The primary partial log-repeat estimate was −0.299 SD (SE 0.174, p = 0.071), and the partial time estimate was +0.015 SD/min (SE 0.094, p = 0.87). The repeat test had power 0.41. A secondary late-window repeat estimate was −0.363 (p = 0.033; not corrected). The repeat-only model fit better than the linear-time-only model, but a decrease proportional to log(1 + time) could not be distinguished from log repeat. In sub-W3, HFA did not decrease over 6.5 minutes (−0.012 SD/min, SE 0.022, p = 0.59). Evidence points to repetition but is not decisive: the planned test did not separate the causes in one patient. The control is a different patient with different contacts and task run, so it provides supporting evidence only. The analysis is limited to one repeat patient and 13 responsive pairs on two adjacent shafts; it assumes residual exchangeability across trials, and the matched-time contrast uses one trial.

speech · high-frequency activity · stereo-EEG · repetition · time in session · TIMIT · permutation test · OpenNeuro