Voice taxonomy detects cognitive decline and Alzheimer markers

In 1,479 human participants across three cohorts, researchers evaluated a new speech framework to detect cognitive impairment, according to a preprint on medRxiv. The team created the Neurocognitive Speech Taxonomy, which organizes 314 voice features into seven neurocognitive domains. These domains include articulatory precision, cognitive-linguistic abilities, executive fluency and planning, phonation and laryngeal control, prosodic modulation, lexical-semantic complexity, and morphosyntactic complexity. The taxonomy showed stable structure across cohorts, with Mantel r values ranging from 0.71 to 0.77. It differentiated mild cognitive impairment with an effect size of d = −0.39 and distinguished Alzheimer's disease biomarker status with an effect size of d = −0.23. The scores also tracked longitudinal cognitive change and correlated with hippocampal volume, with correlation coefficients reaching up to 0.28.