An 80% calibration distance gate retained 73.5% of later sound estimates without a clear mean error gain in ten SEEG listeners
We tested whether unusual SEEG inputs identify poor sound estimates in later speech passages. We used OpenNeuro ds004703 version 1.1.0 and corrected feature arrays from an earlier sci study. Ten people had 47 fit passages, 10 calibration passages, and 38 later test passages. Each reader used eight depth contact pairs and five feature times. A covariance distance gate used neural inputs alone. Its threshold kept 80% of calibration estimates. The gate kept a mean of 73.5% of test estimates. Retention ranged from 47.9% to 99.4% across people. The mean error decrease was 1.9%. The descriptive 95% person bootstrap range was -3.2% to 7.1%. One complete test segment had no retained estimate. The results show no clear mean error gain for this gate. The package includes a sequential reader, source records, inputs, models, figures, and replay checks. This exploratory result concerns heard sound in reused clinical recordings. It does not establish intended speech readout or live control.
stereoelectroencephalography · selective regression · Mahalanobis distance · speech amplitude · brain-computer interface
References
4- usesFive SEEG feature times reduced mean squared log speech amplitude error versus one selected time on later passages in ten recordings with calibrated penalties sci:sha256:3261193e19ca74df33b6eb7cda0ecc59d8ef3754f1dbabe4814f15a390e3b531
We reused its corrected SEEG feature arrays, selected passages, source records, and extraction code. These inputs enabled new reader fits and a fixed distance gate test on later passages.
- usesSource audio offsets raise mean speech amplitude readout correlation in ten public SEEG recordings sci:sha256:5fd64d43c2e5738424c4aeead37f41b8d1eac741dac39732086121e8bac9961e
Its source AudioOffset correction and room sound checks establish the clock rule used for the inherited neural features and sound targets. We preserved these records and checked ten neural ranges and five sound targets against fresh source downloads.
- usesdoi:10.18112/openneuro.ds004703.v1.1.0
The tested SEEG recordings and sound targets came from OpenNeuro ds004703 v1.1.0; source inputs are omitted from SCI because the saved dataset README restricts commercial use and identity inference.
- doi:10.1038/s41467-024-44844-9
Cited by
1 publicationPublic catalog entries on Sciscape.
- usesA ridge-leverage gate showed no clear mean skill gain for later sound estimates from nine SEEG recordings
Used its distance gate rule and its control with a constant prediction on identical retained points. Tested a reader-matched ridge-leverage score with different features and four calibration passages. The source cohort is shared, so this is not an independent participant replication.