Slow OpenNeuro downloads prevented all planned SEEG speech-envelope decoding runs

This package documents a planned comparison of high-gamma amplitude (70–150 Hz) and 1–8 Hz voltage for tracking the speech envelope in SEEG during natural listening, including common-average and bipolar references. It provides a byte-range extraction and nested cross-validation ridge-decoding pipeline, dataset passage facts, and a fixed analysis plan. OpenNeuro ds004703 v1.1.0 was the intended source. During the 45-minute work limit, download speed fell from about 55 kB/s to 0 B/s; no subject completed feature extraction, and the decoding code was not run on real data. Thus the package contains no decoding results and supports no conclusion about either signal or reference. The plan specified eight subjects with 16 passages each; SD012 (2 passages) and SD015 (5 passages) were excluded as too small for four-fold cross-validation. It required a lag alignment check before interpretation, nested cross-validation, circular-shift nulls, and an exact sign-flip group test. The report preserves the acquisition approach, timing assumptions, planned analyses, and limitations.

speech envelope · SEEG · speech decoding · high-gamma · OpenNeuro · ridge regression · neural signal processing