Early pair selection gave a 0.27 dB mean later speech to pause power contrast in seven SEEG patients with supplied audio times

We tested whether SEEG contact pairs selected for greater speech power in two early passages kept that contrast in two later passages. We used seven patients, 28 English passages, and 908 adjacent pairs from OpenNeuro ds004703 version 1.1.0. The fixed 70 to 110 Hz band contained five bins from 72 to 104 Hz. We selected the highest 10% of pairs in each patient. The mean speech minus pause log power contrast fell from 2.03 dB in selection data to 0.27 dB in later data. Five patient estimates were positive. The exact one-sided patient sign-change p value was 0.1016. The 999 circular-shift controls, with pair selection repeated in every draw, gave p = 0.060. The 95% patient bootstrap interval was -0.00044 to 0.57783 dB. The result did not clearly establish a positive group response in later passages. Eleven of 14 independent audio checks passed. The five patients with both checks passed had a mean of 0.32 dB. The main result assumes the supplied audio times. Few pure pauses, a small clinical sample, and reference effects limit the inference. The package preserves scientific results, figures, code, source provenance, and repeat instructions. Source recordings, annotations, and derived input caches are omitted because saved source terms conflict; this prevents a direct offline repeat from the SCI files.

stereoelectroencephalography · speech · high gamma power · held-out test · site selection · neuroscience · OpenNeuro