The speech envelope predicted held-out SEEG high-gamma amplitude better than peak rate in seven listeners

We compared whether speech envelope (ENV), peak-rate acoustic edges (PR), or the half-wave rectified envelope derivative (DER) predicted 70–150 Hz SEEG high-gamma amplitude during passive listening. We analyzed the first 8 of 16 passages (about 4–5 min) for each of 8 OpenNeuro ds004703 subjects, using forward ridge models at 32 Hz with leave-one-passage-out cross-validation. Seven subjects had responsive channels for each reference; SD017 had no responsive common-average-reference (CAR) channel, and SD019 had no responsive bipolar channel. Across the 7 testable subjects, PR minus ENV held-out correlation was −0.025 for CAR and −0.021 for bipolar reference; exact two-sided sign-flip p = 0.016 for each. The unique variance of PR was also smaller than that of ENV (CAR: −0.0046, p = 0.016; bipolar: −0.0050, p = 0.016). DER was worse than ENV (CAR: −0.014, p = 0.016; bipolar: −0.015, p = 0.031). Thus the planned hypotheses that PR would outperform ENV were not supported; the significant effects were in the opposite direction. Adding PR to ENV produced nearly the same CAR mean correlation as ENV alone (0.104 vs 0.103). These findings are limited to 8 passages, one passive-listening task, linear models, one PR definition, and signals smoothed and sampled at 32 Hz; this smoothing may favor ENV. No anatomy-based STG analysis was possible.

speech envelope · peak rate · SEEG · high gamma · speech encoding · ridge regression · passive listening