Pre-onset high-gamma did not identify first-sound articulation place in four Dutch SEEG participants

Question: does SEEG high-gamma (70-170 Hz) in the 300 ms before voice onset identify the place of articulation (labial, coronal, dorsal) of the first sound of a spoken Dutch word? Data: archived features of 4 participants (sub-04, 05, 06, 08) of the public Dutch word-production SEEG dataset, from sci b141a85d and b158d572. Method: plan written before results; labels from spelling; per-participant shrinkage LDA on PCA-20, 5-fold CV x 20, 1000 label permutations. Result: pre-onset high-gamma was not above the null in any participant (balanced accuracy 0.340-0.364, chance 0.333, p 0.24-0.42, Fisher p = 0.35). Post-onset high-gamma was also not above the null (Fisher p = 0.27). The audio after onset identified the class in 4 of 4 (0.50-0.59, p <= 0.005), so the labels are valid. Pre-onset audio was weakly above the null (Fisher p = 0.005), so the pre-onset window was not fully silent. Limits: about 75 trials per participant, only 5-6 dorsal trials, contacts not selected by anatomy; the test detects only effects with balanced accuracy of about 0.42 or more. A speech BCI that must decode articulation before sound will need more trials, contacts in speech motor cortex, or other methods.

SEEG · high-gamma · speech production · place of articulation · Dutch · balanced accuracy · permutation test · speech brain-computer interface