In SEEG, high gamma tracked the speech envelope better than peakRate in 6 of 6 responsive listeners

This study tested whether SEEG high gamma (70–150 Hz) tracked acoustic peakRate (the half-wave rectified speech-envelope derivative) better than the envelope itself. For eight listeners in OpenNeuro ds004703, we analyzed the first six or eight passages per subject (3.2–4.7 min) using lagged ridge encoding models (0–500 ms) and nested leave-one-passage-out cross-validation. Six subjects had 120 contacts that were responsive to the combined envelope-plus-peakRate model under a 200-shift null test (p ≤ 0.01); SD013 and SD017 had none. Across the six responsive subjects, peakRate minus envelope held-out correlation was −0.010 (0/6 positive; exact sign-flip p = 0.031). The signed derivative performed worse (−0.058, p = 0.031). Adding peakRate to the envelope gave no reliable gain (+0.002, 5/6 positive, p = 0.156), and discrete peakRate events did not outperform continuous peakRate (−0.012, p = 0.062). Median held-out r across responsive contacts was 0.085 for the envelope, 0.085 for envelope plus peakRate, and 0.080 for peakRate. These findings support the envelope as a decoding target for the SEEG coverage in this dataset. The study used only 6–8 passages per subject, had no anatomical labels, used an envelope low-pass filtered at 8 Hz and sampled at 32 Hz, and may have had an alignment error of about 0.1 s. The analyzed dataset release is uncertain; source URLs and ETags are retained, while source recordings, audio, event tables, and extracted input caches are not included.

SEEG · high gamma · speech envelope · peakRate · neural encoding · ridge regression · speech brain-computer interface · OpenNeuro ds004703