Table · dataset · 2026

<p>Subgroup analysis: paired contrasts vs. Text-only on the clinically ambiguous subgroup (short-speech <0.89 s non-Healthy, <i>n</i> = 77 clips from 13 participants). Fold-level paired <i>t</i>-tests on 18 of 25 GroupKFold folds (per the pre-specified small-stra<i>t</i>um exclusion criterion defined in Methods). Only Audio+Text shows a statistically detectable advantage over Text-only at . A sensitivity analysis replacing the 7 excluded folds with and re-testing on all 25 folds yields <i>t</i>₂₄ = 2.26, <i>p</i> = 0.033 for Audio+Text vs. Text-only, confirming the result is not driven by the exclusion. This subgroup contrast is exploratory and uncorrected; see Discussion (<i>Statistical multiplicity and the post-hoc nature of subgroup contrasts</i>) for the multiplicity caveat.</p>

Listed in figshare

<p>Subgroup analysis: paired contrasts vs. Text-only on the clinically ambiguous subgroup (short-speech <0.89 s non-Healthy, <i>n</i> = 77 clips from 13 participants).