Readouts from eight SEEG pairs fitted to early passages gave 0.012 mean speech-envelope correlation in later 12-second segments from ten people

We tested whether a readout using eight adjacent-pair SEEG channels, selected and fitted on earlier passages, could estimate the sound envelope in later 12-second speech segments from the same person. The ten-person mean test correlation was 0.012 (descriptive 95% person-bootstrap interval, -0.027 to 0.052); the group time-shift test gave p = 0.273. Nine of ten combined models had greater squared error than a constant training-mean readout, and only one had lower error. The median test correlation was 0.005, mean squared-error reduction was -13.8%, and mean correlations were 0.003 for voltage-only and 0.017 for high-gamma-only ablations. The result applies to this fixed readout, short calibration, and later passages in the same cohort. It does not establish speech content or intent, performance on a new person, day, or live stream, or that eight pairs are intrinsically insufficient. The contribution retains the full person-level results, controls, figures, and model outputs. Source recordings, audio, event tables, and source-derived input caches are omitted because the saved CC0 dataset metadata conflicts with the dataset README's noncommercial and no-identification restrictions; the source dataset and its terms are linked in the scientific text.

stereoelectroencephalography · speech envelope · neural decoding · temporal transfer · brain-computer interfaces · ridge regression · intracranial recordings