A 400 ms delay raised mean speech amplitude correlation in ten offline SEEG readers with eight pairs

We tested whether a 400 ms wait improved a reader of heard speech amplitude. Public OpenNeuro ds004703 version 1.1.0 supplied ten people and 95 selected SEEG segments. Each reader used eight bipolar pairs and 120 inputs. Earlier passages supplied training. The same 4,446 later rows tested all delays. Mean correlation was 0.214 at zero delay and 0.302 at 400 ms. Seven people improved. The mean gain was 0.088. Its descriptive 95 percent person bootstrap interval was 0.014 to 0.157. The exact two-sided sign-flip p value was 0.05664. This reference assumes independent, equally likely signs under a symmetric zero-change null. Mean gains remained when both delays used the same pairs. Removal of three segments with digital clipping gave a gain of 0.074 and an interval across zero. Mean correlation at 100 ms was 0.290. This curve does not establish a best delay. Only five delayed readers beat a constant training-mean prediction by squared error. All 95 SEEG feature segments and 39 audio targets matched fresh source rebuilds. A fresh analysis gave 186 identical files. The package preserves input arrays, source hashes, models, predictions, code, and three figures. This is an exploratory test in reused clinical listening data under the source clock mapping. It does not test intended speech, live control, or brain stimulation. Source recordings, source metadata, and extracted inputs are omitted from SCI because the archived source README restricts commercial use and identity recovery; the reported results remain available, but the package alone cannot rerun the analysis.

SEEG · speech amplitude · brain-computer interface · ridge regression · signal delay · OpenNeuro