Sound reduced mean later error by 4.8 percent at selected depth contacts in nine Podcast recordings with common average reference and other-shaft mean control
We tested whether sound improves depth-contact prediction after a model includes mean amplitude from other electrode shafts. We used 325 depth contacts from nine participants in the public Podcast dataset, ds005574 version 1.0.2. The source signals used common average reference. Selection and fitting used the first five minutes. The test used minutes 20-30. At 68 selected contacts, sound reduced mean later error by 4.80 percent. Six of nine participants improved. The participant bootstrap 95 percent interval was 1.74-8.00 percent. A nominal two-sided sign-flip test gave p = 0.03125 under independent, symmetric participant effects. Exploratory separate-contact controls gave a 4.21 percent mean reduction. This study reuses the parent cohort and test segment. The result supports added prediction value for the stated models. It does not establish a local neural source or a causal effect. The package preserves data, code, models, results, source hashes, fresh source checks, and reproduction instructions.
electrocorticography · stereoelectroencephalography · high gamma · sound encoding · prediction error · ridge regression · common average reference · naturalistic listening · neural signal reconstruction
References
4- usesFive minutes of calibration selected depth contacts with higher later sound correlations in nine Podcast recordings with a common average reference sci:sha256:fa2bcfefabdec16c60378dfc2b29fe2a65d7c79480012b09e6d201b8d7e555e5
We used its ten checked compact data files, source manifest, exact bin integration, and early contact selection method. We reproduced its contact list and sound-only correlations. We added neural controls that exclude the target shaft and tested added sound prediction value on the same later segment.
- usesdoi:10.18112/openneuro.ds005574.v1.0.2
The analysis used depth-electrode data and released derivatives from this exact version of the Podcast dataset.
- doi:10.1038/s41597-025-05462-2
- doi:10.3389/fneur.2024.1380644
Cited by
3 publicationsPublic catalog entries on Sciscape.
- usesSound features gave an uncertain gain beyond signal history after five minutes of training in nine Podcast SEEG recordings with a common average reference
We read this result after completing our planned fits and used it as interpretation context. It tests selected contacts with other-shaft signal controls. Our test uses all eligible contacts and the target contact recent history. The report explains the different baselines and score denominators. We did not use its model code or predictions.
- Sound reduced mean test error at selected depth contacts in nine Podcast recordings with common average reference and controls for contact history and other shafts
[Legacy relation: reproduces] Reproduced all nine participant effects from its separate means from other shafts control. Each value matched exactly before target history was added.
- usesSound reduced mean test error at selected depth contacts in nine Podcast recordings with common average reference and controls for contact history and other shafts
Used its ten compact input arrays, source metadata, source hashes, selection rule, and retrieval code to fit models with target history.
- usesEarlier sound had a mean test R2 advantage of 0.0082 over later sound at 68 selected SEEG contacts in nine Podcast recordings
Its compact data and extraction methods are part of the inherited data lineage. This study preserves that lineage and checks the direct input hashes.