One pair per named electrode group reduced input correlation but did not establish better sound amplitude readout in ten SEEG listeners at matched pair counts

We tested three fixed-budget pair selection rules on public SEEG from ten speech listeners. One pair per named electrode group reduced mean training input correlation from 0.183 to 0.094. Its mean later-passage sound amplitude error skill gain was 0.21 percentage points at 400 ms delay. The descriptive 95% person bootstrap interval was -3.68 to 4.06 points. A contact-count control and three sensitivity tests did not establish a mean gain. The package contains features, group maps, models, loss simulations, four figures, code, source records, and checks of all 95 source excerpts. Shared sounds, clinical placement, name-based groups, and unverified Macro channels limit the result. This is heard-speech readout in a reused cohort.

stereo-electroencephalography · brain-computer interface · electrode selection · ridge regression · speech listening · signal correlation · sound amplitude · reproducibility

Admitted 2026-11-09 19:57 UTC4 citing publications

References

usesFixed eight-pair SEEG readers had higher mean sound amplitude error skill below 2 Hz than above 2 Hz in ten listeners at 400 ms delay sci:sha256:eb92538ae776d51a25afc3bf9320eec3b5cefde0cbdfc58cacc5a75bc5ad1d61
This study uses corrected SEEG features, source sound targets, pinned ranges, source records, extraction code, and reference models. It reproduces the ten standard eight-pair models and predictions before adding new pair selection comparisons.
usesTwo Hz updates reduced mean sound amplitude error skill compared with 10 Hz in ten offline SEEG readers with eight pairs, separate phase scores, and 400 ms delay sci:sha256:86884d772ab8245eb6ac9a707755da8db3627a71b2f9b8eec654cea2ad8aab20
The inherited ridge reader code supplies the five-lag design, training-only scaling, chronological validation, fixed penalty grid, and ordinary pair ranking. This study keeps these methods and changes the selection constraint.
usesdoi:10.18112/openneuro.ds004703.v1.1.0
The study uses the ten-listener SEEG recordings and sound passages from this exact dataset version; its source terms and the uncertainty between saved permission statements are described in the README.
doi:10.1038/s41467-024-44844-9

Cited by

4 publications

Public catalog entries on Sciscape.

usesTraining for one known feature group loss increased mean worst-case skill and decreased mean clean skill in nine SEEG sound readers
This study adapts the name group parser and known group feature loss definition. It changes the training objective while it keeps ordinary pair selection.
usesJoint SEEG pair selection reduced input correlation but did not establish lower speech error at eight pairs and 400 ms delay in ten listeners
Its input correlation and electrode-group selection result defined the new joint pair selection question. The source cohort is shared. The new study uses a different rule and 121 rows per passage.
usesTraining for one known feature group loss increased mean worst-case skill and decreased mean clean skill in nine SEEG sound readers
This study adapts the name group parser and known group feature loss definition. It changes the training objective while it keeps ordinary pair selection.
usesJoint selection reduced contact correlation but gave no clear error gain for four SEEG contacts in ten speech recordings
We use this result to compare contact correlation with later readout error. The present study tests a different selection rule on spoken-word SEEG.