Site ranks gave mean speech versus animal AUC above 0.5 after training without speech in 20 iEEG patients and 20 fMRI adults
We tested whether response order across auditory sites can support transfer between sound families. Separate public studies supplied human iEEG local field potentials and fMRI BOLD estimates. A rank transform removed all response magnitudes within each sound. Models learned voices without speech versus other sounds, then scored speech versus animal sounds. Four electrical patients supplied development. The fixed test reserved 20 other patients and all 20 fMRI adults. Mean rank AUC was 0.76287 and 0.97176. The 95% person Student intervals were 0.67154 to 0.85419 and 0.94402 to 0.99951. Scores exceeded 0.5 in 18 and 20 people. A shared test required evidence in each modality and used common test-label references. It passed correction for two claims, Holm p=0.0002. A fixed mean loss limit of 0.05 AUC failed, Holm p=0.96230. Mean losses from full-value readers were 0.10588 and 0.01441. Five-site panels retained positive mean transfer. Early-area-only samples were uncertain. All negative results and an initial numerical tie audit remain. A separate solver, source checks, and stable tie checks support the calculations. This is a new endpoint test in reused studies, not new acquisition replication. Different cohorts, sounds, source selection, site coverage, references and measurement filters limit scope. Rank invariance is a readout property. No non-acoustic code, shared neural cause, or causal effect is established. The package gives code, permitted electrical inputs, source versions and terms, all predictions and checks, figures, and repeat instructions.
human neuroscience · intracranial EEG · fMRI · auditory cortex · voice · cross-family transfer · spatial ranks · reproducibility
References
11- usesFor two fixed sound panels, neural models trained without speech ranked speech above animal sounds in 20 iEEG patients and 20 fMRI adults sci:sha256:a475145e49c548c9f12b7592bb0fbc3fca3399d536d7b26272b238b0fa65635a
We reuse its permitted electrical means, labels, source receipts and MIT ridge, source retrieval and extraction code. Its full-value transfer result motivated this new rank-only endpoint.
- supportsCalibrated activity ranks separated faces and houses in selected recordings from 7 ECoG patients and 6 fMRI participants sci:sha256:20c499400a21dde9dc8101425fd7864de57292f95847794c163b469b8312ca04
Prior work tested spatial ranks in human visual ECoG and fMRI. We extend the tested property to auditory sound-family transfer, without claiming a new rank transform.
- One-bit face–house readers lost 0.064 AUC in one of six ECoG patients and tied a bit-count reader in one of six fMRI participants sci:sha256:3777d8650b441945fd3b19fc56454befefd94c9fc0be81088ce5b175d1f5e360
Related study of response precision and fixed AUC loss limits in human visual recordings.
- usesA fixed test did not confirm shared voice label gains in 20 iEEG patients and 20 fMRI adults with linear acoustic models sci:sha256:ae42aba7a4f87b32ba1605d8f9711109760fd08a14aec6db9b7787abf51a0df2
Source extraction lineage retained by the reused verified SCI package.
- usesdoi:10.5281/zenodo.20599001
Hect et al. version 1.1 public CC BY 4.0 LFP means, patient and parcel labels, and behavioral sound ordering.
- usesdoi:10.1016/j.cub.2026.07.063
Electrical source study and its voice and acoustic context.
- usesdoi:10.1152/jn.00588.2020
B2021 separate fMRI cohort.
- usesdoi:10.1371/journal.pbio.3002366
Author archive supplies B2021 site responses, participant metadata and numeric sound IDs. Exact source bytes remain external.
- usesdoi:10.1016/j.cub.2022.01.069
Fixed author repository supplies sound names and category metadata used by the inherited extractor.
- doi:10.1038/35002078
- doi:10.1371/journal.pbio.3001675
Cited by
5 publicationsPublic catalog entries on Sciscape.
- usesFor two fixed sound panels, rank readers ordered speech above animal sounds and animal sounds above controls on average in iEEG and fMRI
We reuse its licensed compact electrical data, source readers, extraction receipts and class-balanced ridge method. We test a new lower boundary against independent control sounds.
- usesIn selected panels, rank readers kept mean speech versus animal AUC above 0.5 after each parcel omission in 15 iEEG patients and 20 fMRI adults
Reuse permitted electrical means, aligned sound labels, source provenance and MIT readers. The prior intact-rank and random-panel results motivate the exhaustive worst-parcel endpoint.
- usesMean AUC lower bounds exceeded 0.5 for site errors capped at 5% of training response SD in 20 iEEG patients and 20 fMRI adults
We reuse its permitted electrical means, sound partitions, auditory-site rule and MIT reader and source retrieval code. We add bounded site-error certificates.
- usesA clip correction reduced mean area estimates for three sound groups in 20 iEEG patients and 16 fMRI adults; fixed tests failed
Licensed compact electrical inputs, source labels, retrieval and mapping code, source receipts and audit code.
- usesOne bit per auditory site gave mean four-class accuracy above chance in 20 iEEG patients and 16 fMRI adults
Its auditory site-rank result motivates the stricter fixed-count binary code test. We test a new four-class endpoint on held-out sounds.