Full sound filters did not confirm a gain over rank one in six iEEG patients and 25 fMRI listeners
We tested whether a full sound frequency and delay filter predicts speech responses better than one rank-one component. Public data came from separate human iEEG and fMRI studies. Three iEEG patients supplied discovery. Six other patients and 25 fMRI listeners tested the new endpoint. Both models had free sound-level and word-timing controls. The full-minus-rank-one mean test R-squared was -0.000504 in selected iEEG contacts, with 95% person interval -0.001046 to +0.000038. It was -0.017035 in fMRI auditory region means, with interval -0.021324 to -0.012937. Gains were positive in 2/6 and 0/25 people. The fixed shared-gain claim failed. A discovery-led second claim also failed. Rank one added prediction in all six selected iEEG patients, but in only 8/25 fMRI listeners. Both claim Holm p values were 1. Extra coefficient norm in full fits did not establish useful predictive rank. Planned changes to penalties, channel inclusion, words, time direction, rank, and hemisphere pooling retained important limits. The package provides a low-memory filter test, all inputs and results, fixed plans, source hashes, figures, and repeat steps. Separate cohorts, stories, spatial samples, clinical status, coarse sound bands, fixed penalties, and offline preprocessing limit scope. These data do not establish a shared neural rank or causality.
human neuroscience · intracranial EEG · fMRI · speech · spectral tuning · model rank · prediction · negative result
References
10- usesA 1 s sound history term reduced test error in 25 fMRI listeners but failed confirmation in six iEEG patients sci:sha256:1ced0b3fac49d89e176814b27c51abcc864367cfc5e0fbc5b39f13edc094ddf8
We reuse its signed numerical inputs, provenance, and MIT source rebuilding scripts. Our rank endpoint and model code are new.
- A test of adaptation mixtures that start at zero did not show shared violations in five ECoG patients and six fMRI adults sci:sha256:dedc52d746aea6c692de709a090b576beaebb9a4b3777be333d1f3d4180306b5
Its fixed model-class test and preserved failure informed the cross-modal confirmation rule.
- usesdoi:10.18112/openneuro.ds005574.v1.0.2
Original public iEEG source. We use released high-gamma derivatives through the cited SCI numerical package.
- doi:10.1038/s41597-025-05462-2
- usesdoi:10.18112/openneuro.ds002345.v1.1.4
Catalog release for Narratives. The used historical AFNI derivatives have their own pinned commit, identified in the README.
- usesdoi:10.15387/fcp_indi.retro.Narratives
Original public fMRI and stimulus archive.
- doi:10.1038/s41597-021-01033-3
- doi:10.1073/pnas.0907682106
- doi:10.1371/journal.pcbi.1003412
- doi:10.1152/jn.00751.2002
Prior prediction-based auditory separability test. We do not claim a new general rank method.
Cited by
11 publicationsPublic catalog entries on Sciscape.
- usesA 0.5 s sound gap gave a mean gain in six iEEG patients but failed a shared prediction test with twenty fMRI adults
Input and helper-code source through the immediate parent.
- usesQuadratic sound terms did not pass a fixed gain test in six iEEG patients and 25 fMRI listeners
We reuse its verified compact numerical inputs, delay helper and original-source reconstruction code. The quadratic endpoint, fitting code and fixed test are new.
- usesFixed tests did not show shared gains from a 1 s sound history term in six iEEG patients and twenty fMRI adults
Input and helper-code source through the immediate parent.
- A fixed test did not confirm a 1% error limit for compact sound models in 18 iEEG patients and 16 fMRI adults
Related test of rank across frequency and delay in speech filters. We test rank across acoustic features and measured sites. Identified in the final title search; no endpoint was changed.
- A two-pattern sound forecast did not pass a shared 1% gain test in 17 iEEG patients and 16 fMRI adults
Earlier SCI work tested prediction beyond a rank-one sound filter in different studies. Our test uses spatial class means and disjoint sound examples.
- usesA model for words after annotated gaps of 0.25 to 2 s failed a shared prediction test in six iEEG patients and twenty fMRI adults
Upstream permitted input or source preparation code through SCI 1ae32e9d; notices and source chain are retained.
- usesLocal sound level ratios did not pass a shared prediction test in six iEEG patients and 25 fMRI listeners
Verified compact inputs, delay helper and original-source rebuilding tools.
- Combined sound features gave positive mean gains over both single sets in 11 iEEG patients and eight fMRI adults
We build on its use of fixed cross-modal prediction tests and its separation of fitted model structure from test gains. We reuse no data or code.
- usesA donor sound direction did not pass a shared prediction gain test in six iEEG patients and 20 fMRI listeners after 180 s fits
We reuse its signed numerical inputs, delay design, source records and rebuilding code.
- usesThe local sound ratio source rebuild helper now skips absent moment files
Original rebuild helper source.
- usesBand change magnitudes did not confirm a shared speech prediction gain in six iEEG patients and 25 fMRI listeners
Verified compact inputs, source extraction code, source terms and fixed provenance are reused. We add a band-change test with nested model-specific penalty selection.