A 16-band speech spectrogram did not improve group-level prediction of SEEG high gamma over the envelope during passive listening

We tested whether a 16-band speech spectrogram predicts SEEG high-gamma (70–150 Hz) activity better than the broadband speech envelope during passive listening. In eight listeners from OpenNeuro ds004703, one participant had no envelope-responsive channels, leaving seven for the preregistered group test. The mean held-out correlation gain was +0.0011 (5/7 positive; exact one-sided sign-flip p = 0.438). The STFT sensitivity analysis gave the same conclusion (mean +0.0007, p = 0.430). Across all 1,054 bipolar channels, the spectrogram score was lower in seven of eight participants. At channel level, 13/43 envelope-responsive channels showed a significant passage-level gain, compared with 93/1,011 (9.2%) non-responsive channels. An exploratory analysis performed after the primary result found that gain increased with envelope tracking (Spearman rho = 0.65, n = 43); all 14 channels with r(ENV) >= 0.25 had positive gain (median +0.020), across four of eight participants. These results suggest that most measured channels provided little reliable spectral information beyond the envelope, while strongly envelope-tracking channels showed a small gain. The study used passive listening, 12 passages per participant, and linear ridge temporal response functions; the secondary decoding test in the preregistered plan was not completed. The source dataset has conflicting license statements; recordings, audio, event tables, and source-derived caches are not included.

SEEG · speech · high gamma · speech envelope · spectrogram · temporal response function · brain-computer interface