Sound labels lowered error for omitted acoustic clusters in 18 iEEG patients and two fMRI groups
We tested whether known sound-class labels improve brain-response forecasts when complete acoustic groups are absent from model training. Separate public human studies supplied signed intracranial EEG means and BOLD fMRI estimates. Four patients and four MRI adults supplied development. A local code lock preceded new scores in 18 reserved patients and 16 reserved MRI adults. Mean gain over the selected acoustic model was 6.10 points, with 95% person Student limits 1.56 to 10.64, and 10.75 points, with limits 9.85 to 11.66. One point is 1% of the stated constant-forecast loss. Both modalities also passed an equal-penalty gain check and an absolute-skill check. The single shared rule passed, p=0.007026. Eight adults from another MRI study gave gain 9.44 points. This study shared the MRI sound library. The omitted groups increased median nearest same-class training distance by about 27% over matched random folds. Removing human vocal sounds or using all 88 acoustic descriptors left an electrical gain uncertain. Pre-event electrical data had a positive gain against one model but failed the equal-penalty and skill checks. A separate solver rebuilt all 42 main records within 1.12e-15 in risk. The package supplies code, permitted inputs, source pins and terms, all results, figures and repeat steps. The finding is a prediction result for these processed panels and models. Separate cohorts, prior data reuse, source selection, omitted acoustics and measurement filters limit scope. No uniquely semantic or causal neural process was established.
human neuroscience · iEEG · fMRI · auditory cortex · sound classes · acoustic clusters · kernel ridge regression · prediction · reproducible benchmark
References
14- usesAt 20 calibration sounds, mean class gains over acoustic neighbors were larger in association than early auditory parcels in 14 iEEG patients and 16 fMRI adults sci:sha256:ff51441a87d6dd1f0c2567c840e07fbb6d344674d061956664d946ef0266b5be
Licensed compact electrical data, acoustic measurements, source crosswalks and source-reader structure; prior random-calibration class test.
- usesSound labels lowered test error for averaged responses in 20 iEEG patients and 16 fMRI adults; separate BOLD fits gave no clear gain sci:sha256:495812fc27afe8b3e0fe67db013b2f4b72cecb708745c383879ab0479511bcad
Prior nonlinear acoustic-plus-class prediction design and its limits; the present study holds out complete acoustic clusters.
- usesdoi:10.5281/zenodo.20599001
Hect public data release 1.1, CC BY 4.0; fresh metadata, descriptor and compressed neural-range checks.
- usesdoi:10.1016/j.cub.2026.07.063
Original clinical human iEEG study.
- usesdoi:10.1152/jn.00588.2020
B2021 human fMRI source study.
- usesdoi:10.1016/j.neuron.2015.11.035
NH2015 human fMRI study and sound library.
- usesdoi:10.1371/journal.pbio.3002366
Exact public MRI neural archive and source processing description.
- usesdoi:10.1016/j.cub.2022.01.069
Pinned author source sound-category reference.
- usesdoi:10.1038/nature18933
The six named auditory cortical parcels.
- usesdoi:10.1109/TAFFC.2015.2457417
GeMAPS-family acoustic descriptor definitions.
- usesdoi:10.1145/1873951.1874246
openSMILE acoustic measurements.
- usesMaps fit to other sound classes predicted fixed animal sound patterns in 20 iEEG patients and 16 fMRI adults sci:sha256:1b8693e8389d334be8780044f5056e91081fd0c71bf35ea03565a0b417773b48
Licensed compact data and source-audit donor chain.
- usesOne bit per auditory site gave mean four-class accuracy above chance in 20 iEEG patients and 16 fMRI adults sci:sha256:640399d48b2742bcc4d3c828910fda08a928b409a54c25a630f3a8b53642c0c2
Electrical compact extraction, source chunk pins and MIT HTTP range reader.
- usesTwo fixed acoustic rules did not confirm shared nonvoice prediction gains in 18 SEEG patients and two fMRI groups sci:sha256:f6ade96a55abd5dd6909603449b3d63ad197279407aa009d924a3be21ddc98ae
Licensed MRI acoustic measurements and optional openSMILE rebuild wrappers.
Cited by
4 publicationsPublic catalog entries on Sciscape.
- usesIn fixed sound sets, full class fits reduced mean prediction error beyond one fitted direction in iEEG and two fMRI studies
Licensed electrical means, baseline data, source index maps, class crosswalks and source readers; prior category prediction result in these reused cohorts.
- Vocal labels failed shared forecast tests in 14 iEEG patients and 16 fMRI adults
Related class-label prediction with acoustic groups omitted.
- usesAt 60 sounds, allocation transfer raised mean sampling risk in fixed tables from 18 iEEG patients and 16 fMRI adults
We reuse its licensed compact inputs, class crosswalk, MRI reader and source audit code, with attribution.
- usesFor fixed sound panels, class labels lowered mean error beyond acoustic class scores in iEEG and fMRI
We reuse licensed compact electrical inputs, acoustic tables, source pins and source-reader structure. We test a new training-label acoustic representation and withhold target sounds and people.