In later passages from ten listeners, training on 2 to 5 Hz sound-level changes did not show a clear SEEG readout gain

We tested whether training on 2 to 5 Hz sound-level changes improves a causal SEEG reader of heard speech. We used 95 public passages from ten clinical listeners in OpenNeuro ds004703 version 1.1.0. The earliest 60% of selected passages trained each reader. The later 40% tested it at 400 ms added delay. The mean fast-band skill gain was 0.0263. The descriptive 95% person bootstrap interval was -0.0739 to 0.1494. The exact two-sided person sign-flip p value was 0.5996. Six people had a gain and four had a loss. The original full-target reader had mean skill of 0.1395 below 2 Hz and -0.0374 from 2 to 5 Hz. Fast-band training reduced mean slow-band skill to -0.1238. The new method did not show a clear fast-band gain in these passages. The interval allows both gain and loss. These passages were used in earlier work. This is a new analysis of a known test set. The package preserves compact inputs, code, fitted readers, source range hashes, tables, two figures, and repeat instructions. The fixed analysis was repeated. All 76 analysis files matched by content. A fresh public-source check rebuilt one passage for each person and matched all 25 compact files. The result applies to this offline reader of heard sound. It does not establish performance for intended speech or brain stimulation.

stereoelectroencephalography · speech envelope · brain-computer interface · causal readout · frequency-band error · ridge regression · held-out passages