Eight selected SEEG contacts lowered log speech amplitude error beyond cue and text controls in ten recordings with repaired time stamps

We tested whether eight stereoelectroencephalography (SEEG) contacts add value beyond a known speech prompt. We used ten public Dutch speech recordings from NEMAR nm000371 v1.0.0. The target was the natural logarithm of microphone root mean square amplitude during overt speech. Controls used cue time, log character count, and number status. The final test used 194 later trials with words absent from each person's fitting data. SEEG lowered the remaining mean squared error. The mean per-person decrease was 8.69%. The 95% person bootstrap interval was 5.10% to 12.84%. All ten recordings improved. Text alone lowered error by a mean of 16.38% beyond cue time. Late phase error increased in five recordings. This exploratory test reuses a published cohort and its features. Fresh raw source checks reproduced the feature arrays for 10 of ten recordings. The package preserves code, hashes, features, model states, predictions, source recipes, and figures. Repaired time stamps do not prove physical clock accuracy. This result does not establish intended speech, silent speech, live use, or stimulation benefit.

stereoelectroencephalography · speech production · speech amplitude · brain-computer interfaces · ridge regression · Dutch speech · text controls