Broad spectral structure explained the shift-search excess; local contrast found a candidate ridge in sub-10

We reanalyzed saved audio-by-SEEG correlation matrices from four participants in the SingleWordProductionDutch-iBIDS dataset to assess whether an earlier pitch-shift search detected voice leakage or broad spectral structure. Applying the earlier statistic and frequency-permutation null to gap-aware smoothed matrices gave similarly small p values for the top contacts in sub-04 (0.005 raw; 0.005 smoothed) and sub-05 (0.001; 0.001); raw and smoothed profiles correlated at r = 0.96–0.998 across eight matrices. This indicates that the earlier excess in those participants came from broad spectral structure, for which that null is not valid. A local ridge-minus-flank contrast with a residual-permutation null was significant in sub-10 (top contact and contact mean p = 0.001, with a maximum in the 1–3 semitone band) and in the sub-08 top contact (p = 0.016). Sub-04's contact mean had p = 0.008 over all shifts but p = 0.42 in the 1–3 semitone band; sub-05 had p ≥ 0.06. The sub-10 feature is a candidate, not proof of leakage: the residual null assumes exchangeable residuals across SEEG frequency. The planned time-rotation null on full data, which preserves both signals' spectral structure, did not run because OSF transfer speed was low. The analysis used saved matrices for four of ten participants; the decisive full-data test remains outstanding. Shift signs and exact candidate shifts are withheld because shift values anonymize the voice.

SEEG · speech · acoustic contamination · spectral correlation · null calibration · pitch shift