AI speech analysis of nearly 3,000 Spanish-speaking adults suggests the way people talk can reveal signs of accelerated aging and higher risk of cognitive decline, according to a study published in Science Advances. Researchers found that when an algorithm estimated a speaker’s age as older than their chronological age, those individuals were more likely to have mild cognitive impairment or dementia and to show biological markers of faster aging.
The team assembled voice recordings through the ReD-Lat consortium, testing adults from Argentina, Chile, Colombia, Mexico and Peru. Participants completed a battery of seven brief speech tasks, producing samples that were transcribed and analyzed for more than 700 features such as pauses, rate, pitch, vocabulary and emotional tone. A machine-learning model was trained to predict chronological age from those acoustic and linguistic features, and the difference between predicted and actual age was reported as a “speech-age gap.”
Analysis showed systematically larger speech-age gaps among people with cognitive diagnoses, with the greatest discrepancies in language-predominant forms of dementia. Larger gaps also correlated with poorer performance on memory, language and attention tests, greater difficulty with daily activities, and adverse results on epigenetic aging measures. Among participants with Alzheimer’s disease, the speech-age gap was associated with higher levels of the protein p-tau217. Social and economic hardships—financial strain, food insecurity, limited healthcare and lower education—were more common in those whose voices sounded older.
Investigators and outside clinicians stress limitations: many participants were assessed once, preventing conclusions about whether older-sounding speech predicts future decline, and the model was developed and tested only in Spanish. Clinicians note that voice can change for many reasons, including mood and stress, so the tool is not a standalone diagnostic test. The researchers plan longitudinal studies and validation in other languages to determine whether speech markers can reliably signal impending cognitive decline and be adapted for broader clinical screening.
