For 165 fixed sounds, four-component models reduced repeat error in six SEEG patients and 24 fMRI adults
We tested one-component and four-component models of repeated responses to 165 natural sounds. Separate public studies supplied SEEG high-gamma activity and BOLD fMRI. We removed three broad class means and each sound's mean across sites. One block fit the patterns. A second block set their weights. A third block tested prediction error. Development used two patients and four adults. A local code lock preceded tests in six other patients and 16 other adults. A fixed follow-up used eight adults from a separate fMRI study. Mean added gains were 2.742%, 2.866% and 1.535% of residual test energy. Pointwise 95% person intervals were 1.022-4.463%, 2.037-3.695% and 0.468-2.601%. All 30 reserved people had positive gains. The main and extended shared tests each passed a two-claim Holm rule, p=0.03125. Sixteen-site checks kept positive mean gains. Nonspeech voice intervals crossed zero in the two main cohorts. An exact model with one true signal pattern and independent structured noise gave a 10% gain in every draw. Thus, a gain over the first empirical component does not establish latent signal rank. This is a new prediction test in reused cohorts and a tested tool. Source site selection, fixed sounds, different cohorts, measurement filters and possible repeat artifacts limit scope. No common neural cause is established. The package preserves code, source pins and terms, all results, figures, audits and repeat steps. Original neural arrays stay external.
human neuroscience · SEEG · fMRI · auditory cortex · natural sounds · repeat prediction · signal and noise · dimensionality
References
10- usesSpeech and song axes gave positive mean repeat scores for nine voice clips in SEEG and two fMRI cohorts sci:sha256:6720e49a50bda02902607508181e2616c7f5e48b21bceee56a2d3fcdc618bdae
We reuse its MIT source preparation code, exact source pins and anatomical MRI rule. We test a different prediction endpoint.
- usesAfter calibration to unrelated sounds, speech patterns had greater pair agreement than other voices and animal sounds in human iEEG and two fMRI cohorts sci:sha256:91527309e9faffa6262bc6da6a1dc64d6a38c285dc4b2c5df3ad3967434a3159
Its measured concentration result and stated noise limits motivate the repeat-prediction question. We do not reuse its electrical cohort.
- usesdoi:10.5281/zenodo.6544488
Public SEEG data version v1.1; source bytes are fixed by Git commit and SHA-256 values.
- usesdoi:10.1371/journal.pbio.3001675
SEEG source study and processing methods.
- usesdoi:10.1152/jn.00588.2020
B2021 source fMRI study.
- usesdoi:10.1016/j.neuron.2015.11.035
NH2015 source fMRI study and natural-sound panel.
- usesdoi:10.1371/journal.pbio.3002366
Public author archive of repeated fMRI estimates. We pin the tar bytes separately.
- doi:10.1038/s41586-019-1346-5
Prior use of repeat measurements to study neural response dimensions.
- doi:10.1371/journal.pcbi.1012092
Prior work shows why structured noise and signal dimensions must be separated.
- With clipped acoustic inputs, rank 2 models had lower mean test error than rank 1 for fixed panels from 20 iEEG patients and 20 fMRI adults sci:sha256:a7ebe1050fc6e865343e629ad39c22ef72edc352e2f1390afdc35c73a0231229
Closely related acoustic rank comparison and known-rank controls. Its full signed text was read in the final audit. It informs the final scope statement, not the fixed test or its code.
Cited by
22 publicationsPublic catalog entries on Sciscape.
- usesAdded guide patterns reduced test sound response error in SEEG and two fMRI cohorts with source processing
We reuse its MIT source preparation code and exact source pins. We extend its estimator warning to site completion on sounds omitted from fitting.
- usesResponse and repeat difference covariance maps had positive mean agreement for fixed sounds in six SEEG patients and 24 fMRI adults
We reuse its MIT source preparation code unchanged and its pinned public-source manifest.
- usesFor 165 sounds, mean scores for own sound maps exceeded scores for other people in six SEEG patients and two fMRI cohorts
We reuse MIT source preparation code, data pins and provenance for a new record-identity endpoint.
- usesA fixed test of repeat distances did not confirm shared tree rejection in six SEEG patients and 24 fMRI adults
We reuse its MIT source preparation code, source pins and cohort selection. We test a new endpoint and retain source-selection limits.
- usesA diversity rule for eight sites failed a shared sound matching gain test in six SEEG patients and 24 fMRI adults
Original MIT preparation reader and source manifest, obtained unchanged through SCI 4cd19f9d.
- usesIn exploratory tests, acoustic feature clipping lowered baseline error in 20 iEEG patients and 24 fMRI adults; a fixed slope test failed
Retained source preparation or licensed derivative ancestry; no prior outcome is counted as new confirmation.
- usesOne bit per site gave sound match scores above chance in selected panels from six SEEG patients and 24 fMRI adults
We reuse its source pins and MIT preparation code for the three-repeat human sound-response benchmark. Our endpoint tests sign-code sound matching rather than low-rank prediction.
- usesAt 24 selected sites, a fixed test did not confirm a shared response and repeat variation link in SEEG and fMRI
It supplied the earlier preparation code in the preserved donor chain.
- usesAfter site calibration, sound matching without site labels failed a fixed shared test in six SEEG patients and 24 fMRI adults
Pinned sources and MIT preparation code; related repeat-prediction benchmark.
- usesAt 16 selected sites, separate sound sets gave positive covariance shape contrasts in six SEEG patients and 16 fMRI adults
This is the upstream donor for the unchanged repeated-response preparation code.
- Fixed maps from eight sites gave higher prediction gains within a repeat in six SEEG patients and two fMRI studies
Found in the final catalog check after scoring. Prior component prediction of repeated known sounds differs from our held-out-sound, disjoint-site map.
- usesFor 54 sounds, donor panel gains were uncertain in six SEEG patients and positive in 24 fMRI adults
Upstream preparation-code attribution retained through SCI 72b300b9.
- usesAt selected sites, four nonvocal axes captured extra speech-and-song repeat energy in six SEEG patients and 24 fMRI adults
We reuse its MIT source preparation code and exact source pins. We test transfer to a sound group excluded from fitting.
- usesTests of two sound groups did not confirm order crossings in six SEEG patients and 24 fMRI adults
Earlier MIT preparation code lineage through SCI 38426b4d.
- usesAt fixed total weight, nonvoice filters failed a shared speech and song test in six SEEG patients and 24 fMRI adults
Its repeat prediction result motivated the controlled error comparison.
- usesCalibrated extra axes reduced mean repeat error in 16-site panels from six SEEG patients and 24 fMRI adults
Extends its repeat-response benchmark and interpretation limit with disjoint sound roles and a four-axis baseline.
- usesFour low-repeatability guides failed a shared prediction test in six SEEG patients and 16 fMRI adults
Upstream source preparation attribution.
- usesFour linear axes failed a shared prediction test against a scalar curve in human SEEG and fMRI sound responses
Its four-component repeat prediction result motivates complete sound exclusion and a scalar nonlinear competitor. Our endpoint and processing differ.
- usesMean gains from fixed quadratic terms were below one error point in six SEEG patients and 16 fMRI adults
Its repeat-component test informed the fixed rank-four prediction question and the noise interpretation check.
- usesMean sound class scores from repeat spread were above chance in six SEEG patients and 16 fMRI adults; a fixed shared test failed
Upstream preparation code attribution chain.
- usesA fixed auditory response fit did not pass a shared repeat error test in six SEEG patients and 16 fMRI adults
Closest prior auditory four-versus-one repeat-fit benchmark and neural-rank interpretation limit. Reuses its unchanged MIT model for an exploratory baseline sensitivity replay; original six SEEG person scores match exactly.
- usesArea scores from four axes were positive for 165 sounds after class mean removal in six SEEG patients and 24 fMRI adults
We build on its prediction-gain limitation and rank-one noise example. It supplies source pins and MIT preparation code. Our weighted area endpoint differs from its prediction-error endpoint.