In six patients, high-gamma onset did not provide response-specific single-trial speech timing

We tested whether a single-trial high-gamma (HG) onset detector could estimate response time (RT) in auditory naming data from six patients in OpenNeuro ds005007 v1.0.0. The analysis used one run per patient (76–95 trials; 26–53 good contacts) and inferred event codes, defining RT as the response marker minus sound offset. In five-fold cross-validation, response-selected contacts produced a positive rank correlation between detector time and RT in 5/6 patients (significant in 3/6); mean rho was 0.16 and the group sign-flip p was 0.094. The sound-responsive contact control tracked RT in 6/6 patients (mean rho 0.13; p = 0.031). The response-minus-control difference was 0.03 (p = 0.59). A calibrated neural estimate reduced median absolute error relative to a constant guess in 3/6 patients, but the median error ratio was 1.01 (p = 0.69). Miss rates were 0–3%. Thus, the weak correlation was not specific to response-selected contacts, and this detector did not provide usable single-trial timing in these data. Event codes were inferred; the marker-to-voice delay is unknown and no audio channel was available. The study includes six patients and one run each, has low-powered group tests, and tests only one detector and threshold rule. BIDS metadata labels the channels ECOG; electrode type was not confirmed. EDF files for sub-07/sub-08 and sub-09/sub-10 were byte-identical pairs, and subjects 07–12 used an annotation scheme that could not be decoded, so they contributed no trials.

high gamma · speech timing · auditory naming · response time · intracranial electrophysiology · single-trial decoding · control analysis