Time contrasts raised mean sound amplitude skill by 3.54 points over band ratios on 100 later SEEG passages in nine people at 400 ms delay

Time contrasts improved a gain-protected sound amplitude reader in public SEEG. We tested nine people and 100 later speech passages from OpenNeuro ds004703 version 1.1.0. Each reader used eight bipolar pairs and a 400 ms output delay. Adding four temporal contrasts per pair raised mean error skill from 0.0497 to 0.0851. The mean increase was 3.54 percentage points. The descriptive 95% person bootstrap interval was 1.09 to 6.09 points. Six people improved. The exact person sign-flip value was p = 0.05469. Test passages had been used before, so this result is exploratory. An orthogonal projection retains all linear contrasts that cancel one constant log-power shift per pair. Cancellation requires constant positive gain across both bands and the complete input support, with an inactive power floor. Protection cost 4.49 mean skill points relative to full power on the original data. An exact risk formula gives a modeled mean crossing at 0.787 dB shaft gain standard deviation. These gains are stress assumptions, not measured hardware changes. Fresh downloads checked all 154 neural excerpts and 48 sound targets. An independent solver checked 27 readers. A complete replay matched 47 core result files. The package preserves data, provenance, code, fitted models, predictions, tables, four figures, and replay instructions. This study concerns heard sound. It does not test intended speech or brain stimulation.

stereo-EEG · brain-computer interface · sound amplitude · gain invariance · temporal contrasts · ridge regression · exploratory study