Early SEEG high-gamma did not predict speech reaction time in four Dutch word-production participants

We tested whether stereo-electroencephalography (SEEG) high-gamma (70–170 Hz) activity in the first 0.3 s after a Dutch word cue predicts that trial’s speech reaction time (RT). We analysed four of ten participants (sub-01, sub-02, sub-05, sub-06; 95, 97, 94, and 96 trials used). High-gamma features were baseline-corrected means in three 100 ms bins. Ridge regression with 10-fold cross-validation predicted RT after removal of a linear trial trend; significance was assessed with 200 within-participant permutations and an exact one-sided group sign-flip test. No participant had permutation p < 0.05 (Spearman r = −0.271, 0.094, 0.082, 0.048); the group mean r was −0.012 (sign-flip p = 0.56). For raw RT, only sub-05 was significant (r = 0.243, p = 0.010), while RT increased over the session (Spearman r with trial number = 0.39); the detrended result was r = 0.082 (p = 0.085). A pre-cue high-gamma control had mean r = −0.17. Of 279 task-responsive contacts, 155 (56%) had a higher voice-locked than cue-locked template R², but most lock-index values were between −0.01 and +0.01 because the RT interquartile range was 0.12–0.16 s. The high-gamma/audio alignment check reached r = 0.31–0.65 across participants and exposed a 102.4-to-100 Hz alignment error that was corrected for the reported results. This negative pilot does not establish that no effect exists: only four participants were analysed, and the smallest possible group sign-flip p-value was 0.0625. The prediction window was short, the task was simple with a narrow RT spread, voice onsets were estimated automatically from pitch-shifted public audio and were not checked by hand, and 200 permutations give p-value resolution 0.005. The other six participants were not completed within the run time limit. Full methods, results, source hashes, code, and limitations are preserved in the README and original files.

stereo-electroencephalography · high-gamma activity · speech reaction time · brain-computer interface · Dutch word production · ridge regression · permutation test · voice onset