Eight SEEG contacts reduced mean error by 1.60 percent for sound windows that end 200 ms after released input in ten records with cue and past sound controls

We tested speech amplitude forecasts in ten public SEEG records from NEMAR nm000371 v1.0.0. Each reader used eight contacts. The target was a 50 ms sound window that ended 200 ms after the latest released input. The control used cue phase and 250 ms of past sound. On 194 later trials, SEEG reduced the equal-person mean error by 1.60 percent. Error fell in six people. The 95 percent person bootstrap interval was 0.02 to 3.55 percent. The person sign-flip reference gave p = 0.137. The effect was larger early in the trial and negative later. The result is exploratory. Source processing and physical delays do not certify online use. We preserve inputs, source hashes, raw retrieval steps, fitted readers, predictions, five figures, and checks. All ten full source hashes were checked. A fresh raw rebuild matched every published input array.

stereo-EEG · speech amplitude · forecasting · brain-computer interfaces · ridge regression · overt speech