Past audio models had higher SEEG test scores than future audio models in four of five ds004703 patients

This study compared past and future speech amplitude models in public SEEG data from OpenNeuro ds004703 version 1.1.0. Both models included concurrent amplitude and had six columns. Pairs and coefficients were chosen on 20 early passages. Test scores used 23 separate later passages in five patients. Past models had higher R² in four patients. The equal patient mean past-minus-future gain was 0.03978. The planned exact two-sided patient sign flip test gave p = 0.125. It did not meet alpha 0.05. All means after removing one patient were positive. Planned model and amplitude checks kept positive mean gains. Circular-shift controls supported a relation to audio timing under rotation assumptions. They did not test the same patient population null. The package preserves source features, code, predictions, control draws, provenance, independent checks, and retrieval instructions. This secondary analysis applies to the selected pairs, passages, and model. It does not establish a neural cause or an absolute response delay.

stereoelectroencephalography · speech envelope · high gamma · ridge regression · temporal prediction · time direction · OpenNeuro