Neural tracking correlations and null distributions of paper 'Modeling and Interpreting Correlations, Null Distributions and Significance Levels in Neural Tracking of Natural Stimuli'

Le résumé fourni par la source

Precomputed real and null Pearson correlations between EEG responses and eight stimulus features, for 121 participants listening to a single ~876 s Flemish story ("Milan," narrated by Stijn Vranken). This dataset accompanies the paper Geirnaert, Bertrand, Francart & Vanthornhout, "Modeling and Interpreting Correlations, Null Distributions and Significance Levels in Neural Tracking of Natural Stimuli" [1], and is the input required to reproduce every quantitative result in it. The accompanying analysis code is available at https://doi.org/10.5281/zenodo.21821416. The correlations were obtained with a linear backward (decoding) model reconstructing each stimulus feature from 64-channel EEG (BioSemi ActiveTwo), re-referenced to the common average, drift-filtered, and downsampled to 8 Hz. Real (aligned) correlations were computed between each reconstructed and actual feature per window; null (misaligned) correlations were generated by pairing each reconstruction with genuine but temporally unrelated segments of the same feature, per the misalignment method described in the paper. The EEG recordings are part of the publicly available SparrKULee dataset, supplemented here with additional participants whose raw EEG could not be made public; only the derived correlations for those participants are included. File correlations.mat (~2.4 GB), a MATLAB v7.3 (HDF5-based) .mat file containing a single struct data with one field per window length, data.length_ for w ∈ {1, 2, 5, 10, 20, 30, 60} seconds. Each holds: r — real (aligned) correlations, size features × participants × windows. NaN marks a window that could not be evaluated. nullR — null (misaligned) correlations, size features × participants × windows × permutations. Dimensions 121 participants 8 stimulus features, in this fixed order: (1) envelope (broadband, ~0.1–4 Hz), (2) smallband envelope (1–1.1 Hz), (3) acoustic edge, (4) spectral flux, (5) phoneme onset, (6) word onset, (7) word frequency, (8) punctuation onset. 7 window lengths: 1, 2, 5, 10, 20, 30, 60 s (the paper uses 1-20 s). Sampling frequency 8 Hz. Usage This file contains derived correlations only, not raw EEG or audio. Load it in MATLAB with load('correlations.mat'), or use the loader functions provided in the code repository (loadDistributions.m, loadNullDistributions.m), which also document the exact field layout. See the code repository's README for the full mapping from this file to every figure and table in the paper. [1] Geirnaert, S., Bertrand A., Francart T., Vanthornhout J., 2026. Modeling and Interpreting Correlations, Null Distributions and Significance Levels in Neural Tracking of Natural Stimuli. arXiv preprint arXiv:2608.10887. doi: 10.48550/arXiv.2608.10887.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.