A SEEG voltage decoder with eight pairs gave 49.4% mean balanced accuracy for heard vowels and unvoiced fricatives in later English passages from ten people

We tested whether signed voltage time bins improve a small decoder for heard vowels and unvoiced fricatives. We used public OpenNeuro ds004703 version 1.1.0. The sample had ten people, 3,516 sounds, and 1,484 test sounds in later English passages. Each fixed linear model used eight bipolar SEEG pairs. Four voltage bins gave 49.4% mean balanced accuracy. Four power bands gave 47.4%. The paired mean difference was +2.0 percentage points. Its 95% person bootstrap interval was -2.0 to +6.3. The exact two-sided sign-flip p value was 0.404. Before-sound voltage gave 50.3%. These results do not establish an advantage for voltage or useful group decoding in this test. One person reached 62.8%. A later check that removed events with digital clipping gave 58.7% for that person. The SCI package retains code, model decisions, checks, and figures. Restricted source inputs and source range records remain in the local historical archive. The test concerns heard speech at known sound times. It does not test intended speech or brain stimulation.

stereoelectroencephalography · auditory decoding · voltage features · balanced accuracy · heard speech