Fixed speech models had mean squared test error of 0.919 with past and current amplitude and 0.958 with future and current amplitude in five SEEG patients

We tested whether past speech amplitude improved fixed models of selected SEEG high gamma power when both past and future models also included current amplitude. Using OpenNeuro ds004703 version 1.1.0, we selected one contact pair per patient and fitted ridge models on 20 early passages, then tested on 23 later passages from five patients. Mean squared test error was 0.91944 with past and current amplitude and 0.95810 with future and current amplitude. Four of five patient differences favored past amplitude. The mean future-minus-past difference was 0.03866 (95% t interval -0.00115 to 0.07847; exact two-sided sign flip p=0.125). An assumed 40 ms audio shift reduced the difference to 0.01360. This small, non-independent reanalysis did not establish a group difference or neural cause. The original results, figures, predictions, and analysis code are retained. Source recordings, metadata, event records, and extracted input caches are linked but omitted because ds004703 source statements conflict: its metadata says CC0, while its README restricts commercial use and participant identity analysis.

stereoelectroencephalography · speech amplitude · high gamma power · ridge regression · temporal prediction