A 400 ms delay increased mean heard speech log amplitude correlation in ten SEEG readers with equal input size

We tested output delay in ten separate readers of heard speech log amplitude from public OpenNeuro SEEG data. The source was ds004703 version 1.1.0. All delays used the same 38 later test passages, 113 targets per passage, and 120 input columns. Mean passage-centered correlation was 0.2182 at zero delay and 0.3053 at 400 ms. Eight people had an increase and two had a decrease. The paired mean increase was 0.0871. Its descriptive 95% person bootstrap interval was 0.0111 to 0.1540. The exact two-sided sign-flip value was 0.05664 under exchangeable paired signs. The population effect remains uncertain. A same-pair check retained the mean correlation increase. A delay rule set from early passages selected a mean 170 ms delay and gave later correlation of 0.2993. This does not prove equal performance. Absolute error was unstable and included a large failure. These reused passive listening records do not establish a neural signal source, intended speech decoding, live control, or brain stimulation benefit. The package preserves four figures, all models and test predictions, source versions and hashes, a local analysis plan, code, and repeat instructions. Independent timing and model checks, a complete source array rebuild, and a clean numerical replay passed.

stereoelectroencephalography · speech amplitude · output delay · ridge regression · neural signal tracking · reproducibility