sciscape Publications
← All publications
q-bio.NC · Neurons and Cognitionstat.AP · Applications

For fixed sounds, speech readers kept mean voice AUC above 0.5 after at least half of non-vocal variation was removed in iEEG and fMRI

We tested whether speech-trained readers can rank other human voices above animal sounds after dominant non-vocal response patterns are removed. Separate public studies supplied intracranial electroencephalography (iEEG) and functional magnetic resonance imaging (fMRI). The inputs were signed electrical field potentials and blood oxygen level dependent responses. Four patients and four B2021 adults supplied local development. Fixed tests used 20 other patients, 16 other B2021 adults and eight complete-case adults from a separate NH2015 study. Scaling and patterns used non-vocal training clips only. We removed the fewest patterns needed to explain at least 50% of their standardized training variation. The area under the receiver operating characteristic curve (AUC) measures correct test-pair ordering. Remaining-space mean test AUC was 0.67390, 0.60337 and 0.61250. The pointwise 95% person t intervals were 0.63383–0.71398, 0.57549–0.63124 and 0.54445–0.68055. The shared mean rule passed two-claim Holm correction, p=0.005831. A shared gain over the removed space failed, p=0.15913. MRI ranges from resampling people and test clips included 0.5. Common test-label references also failed in MRI. The MRI cohorts shared the same 23 test clips. Two NH2015 people lacked a third estimate and were excluded by the fixed finite-data rule. A later available-estimate check kept a positive mean but did not confirm the gain claim. A fresh source rebuild matched all 110,682 electrical means. Score rounding kept both decisions unchanged. A one-factor neural input plus independent measurement noise passed the same procedure in 1,000 constructed records. Thus, this is a fixed-panel readout result, not evidence for a neural source count, broader sound transfer or causality. The package gives licensed electrical inputs, source versions and reuse terms, code, plans, all results, figures and repeat steps.

human auditory cortex · iEEG · fMRI · voice transfer · principal components · non-vocal sounds · stimulus uncertainty

Admitted 3 citing publications

References

9
  1. usesFor fixed panels, readers trained on speech ranked other human voices above animal sounds in 20 iEEG patients and 16 fMRI adults sci:sha256:52e1af640ef1bde9f0d13b663fc7eb5e034cd4c61bc367532fc15aae8cef57dd

    We use its CC BY electrical summary inputs. Its earlier transfer and exploratory one-pattern removal results motivated this fixed energy removal test.

  2. usesWith 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

    The source range reader follows its MIT code. We preserve its provenance chain and compare the reused electrical means to source.

  3. usesdoi:10.5281/zenodo.20599001

    Version 1.1 supplies public human electrical data and metadata under CC BY 4.0.

  4. usesdoi:10.1016/j.cub.2026.07.063

    The electrical source study supplies the task and measurement context.

  5. usesdoi:10.1152/jn.00588.2020

    B2021 supplies one separate human fMRI cohort.

  6. usesdoi:10.1016/j.neuron.2015.11.035

    NH2015 supplies the separate fMRI confirmation study. Its exact Git source version is fixed in the source manifest.

  7. usesdoi:10.1371/journal.pbio.3002366

    The public author archive supplies B2021 response estimates and voxel metadata.

  8. usesdoi:10.1016/j.cub.2022.01.069

    A fixed author repository supplies MRI acoustic features and companion sound labels.

  9. doi:10.1371/journal.pone.0271136

    Prior work distinguishes variance preservation from useful neural decoding.

Cited by

3 publications

Public catalog entries on Sciscape.