Kragel, J. E.

High-frequency "ripple" oscillations support learning and memory across species, yet it has been argued that putative ripples in awake human recordings are false positives produced when algorithms misread aperiodic (1/f) fluctuations as ripple-band oscillations. We show that this conclusion arises from an artifact of evaluating detection algorithms on surrogate data containing only aperiodic activity. Ripple detectors are adaptive, setting their threshold from the amplitude statistics of the signal, so applying them to surrogate data that contains only aperiodic activity lowers the threshold and inflates false positives (median 62%). Adding real ripple-band events back to the surrogate corrects this threshold shift and eliminates most false detections across multiple standard algorithms. Using multivariate classifiers, we show aperiodic fluctuations can reproduce the power of ripples but not their timing or spectral content. These findings indicate care needs to be taken when using surrogates to evaluate ripple detection algorithms. Thus, under realistic signal properties, human hippocampal ripples remain distinguishable from aperiodic activity.