The 95 percent listener interval for added sound band EEG prediction gain included zero in two ds004408 excerpts under source alignment

We tested whether eight added sound band envelopes improved prediction of standardized EEG file values. Ten listeners supplied fitting data. Nine other listeners supplied test data. Each model predicted a different story excerpt from its fitting excerpt. Mean error gain was -0.0000313. Positive gain favors the added features. The conditional 95 percent listener interval was -0.0001605 to 0.0000953. Thus this test did not establish a mean gain. Two fixed common penalties increased mean error. Longer lags and a longer edge trim gave intervals that included zero. The interval holds fitting listeners and both excerpts fixed. Source time alignment is assumed. Physical voltage calibration is unresolved. The archive preserves 38 processed EEG records, sound features, code, models, predictions, individual scores, source hashes, and repeat checks.

speech EEG · sound band envelopes · EEG prediction · ridge regression · cross-validation · listener bootstrap · OpenNeuro