Phrase onsets add a small, nonsignificant prediction gain beyond the speech envelope in ds004703 SEEG

In passive listening data from OpenNeuro ds004703, we tested whether phrase-onset impulses improve held-out prediction of bipolar SEEG high-gamma amplitude beyond the speech envelope. We analyzed the first 8 passages (about 4.6 min) from each of 8 subjects using ridge encoding models with 0–500 ms lags and leave-one-passage-out validation. Six subjects had responsive channels (41 total). Adding the onset term produced a mean gain of +0.0083 in held-out r, positive in 4 of 6 subjects; the preregistered exact one-sided sign-flip test was not significant (p = 0.0625). The envelope carried most of the predictable signal (r(ENV+ONS) − r(ONS) = +0.112, 6 of 6 subjects, p = 0.0156). In a post hoc control, onsets from other passages gave a mean gain of −0.0027; true onsets gave a larger gain in all 6 subjects (p = 0.0156). The onset kernel peaked at about 60 ms. The findings indicate a small onset-specific response in some channels, while the envelope remains the main linear target. Limits include 8 passages, 6 subjects in group tests, passive listening without behavior, no anatomy-based channel selection, a fixed acoustic-pause detector, a liberal channel-level test, and a post hoc control. The exact dataset release used is uncertain.

speech envelope · phrase onset · SEEG · high gamma · ridge regression · encoding model · passive listening · OpenNeuro ds004703 · speech processing