Bipolar SEEG high-gamma fit the speech envelope better than its positive derivative in 6 of 7 subjects

In passive speech listening, we compared how well 70–150 Hz high-gamma amplitude in bipolar SEEG was fit by the speech envelope (ENV), its half-wave rectified derivative (EDGE), or both. We analyzed the first four passages from eight OpenNeuro ds004703 subjects at 32 Hz, using forward ridge models with 0–500 ms lags and leave-one-passage-out cross-validation. Channels were selected using held-out BOTH-model r > 0.1. From 1 s after passage onset to passage end, mean r_EDGE − r_ENV was −0.034, negative in 6 of 7 subjects (exact two-sided sign-flip p = 0.047; 128 patterns); SD019 had no channel pass selection. The unique-variance comparison was −0.010, negative in 6 of 7 subjects (p = 0.031), favoring ENV. Including onset, the mean difference was −0.020 in 7 of 8 subjects (p = 0.055). These are weak results: only four passages (about 2 min) were analyzed per subject, r values were small, and both predictors had median TRF peak lag 0 ms. Zero-phase filtering, an earlier reported alignment offset, untested negative lags, and the leading EDGE signal may affect the comparison. At this montage and resolution, ENV is a sufficient target for an envelope-based speech BCI; the result does not reject peakRate coding at selected STG sites.

speech envelope · high-gamma · SEEG · speech encoding · temporal response function · brain-computer interface · OpenNeuro ds004703