In one SEEG participant, a high-gamma detector separated pre-response from in-sound windows but did not show response timing

In one participant and one run of an auditory naming task (90 trials, 45 contacts), we assessed whether SEEG high-gamma activity could indicate when a spoken response was about to start. The event codes were inferred from timing: X101/X501 were treated as sound onset, X302/X702 as sound offset, and X401/X801 as response markers. We do not know the codebook or the delay from the response marker to speech. A leave-one-trial-out shrinkage LDA separated the 300 ms before the response marker from 0.6–0.9 s into the sound with AUC 0.90 (p=0.005, 200 label swaps), and from pre-sound windows with AUC 0.79 (p=0.005). Eighteen contacts had a high-gamma increase before the marker; 39 had an increase after it. However, across positive windows moved from 1.5 s before to 0.7 s after the marker, AUC remained 0.83–0.91 (null 95th percentile 0.58); it was already 0.85 for a window ending 1.2 s before the marker. Thus, the high pre-response versus in-sound AUC does not establish a speech-timing signal: it can reflect separation of the post-sound state from the in-sound state. A speech-onset detector for a brain-computer interface should pass a sliding-window control or use negative windows from the silent gap after the sound. The result is limited to one participant and one run, with inferred event codes and unknown marker-to-speech delay; post-sound auditory and pre-speech activity overlap. The package includes the original analysis code, results, tables, figure, and retrieval recipe. The analyzed files came from the public OpenNeuro ds005007 S3 mirror; the exact dataset release is not established.

SEEG · high-gamma activity · speech timing · brain-computer interface · auditory naming · sliding-window control · leave-one-out classification