A machine‑learning system that evaluates hundreds of speech and language features can estimate a person’s chronological age from the way they talk, and a larger discrepancy between predicted and actual age may signal dementia risk and faster biological aging, a study published in Science Advances finds.
How the speech clock works
The research team analyzed data from the ReD‑Lat consortium, a large dementia‑focused project across Latin America. The sample comprised 2,900‑plus Spanish‑speaking adults from Argentina, Chile, Colombia, Mexico and Peru, ranging in age from 18 to 88. Roughly half of the participants were cognitively healthy, while the remainder had mild cognitive impairment, Alzheimer’s disease or frontotemporal dementia that primarily affects language.
Each participant completed seven spoken tasks, including describing an animated video, naming as many vegetables or animals as possible in 60 seconds, and retelling a short story immediately and again after a 20‑ to 30‑minute delay. Recordings were transcribed and more than 700 acoustic and linguistic features—such as pause length, speech rate, pitch, vocabulary richness and emotional tone—were extracted.
A machine‑learning model was trained on these features together with the participants’ known ages, teaching the algorithm to predict age from voice. The resulting “speech‑age gap” was calculated by subtracting a person’s actual age from the AI‑estimated age.
Links between speech‑age gap, cognition and biology
People whose voices sounded older than their chronological age showed the greatest speech‑age gaps. Cognitively healthy individuals exhibited the smallest gaps, whereas those with mild cognitive impairment or dementia displayed progressively larger discrepancies. The widest gaps appeared in participants with language‑dominant frontotemporal dementia.
Larger gaps correlated with poorer performance on standard memory, language and attention tests, as well as reduced ability to manage everyday tasks. Blood‑based epigenetic clocks—three separate measures that estimate biological age from DNA methylation patterns—also showed greater age acceleration in participants with higher speech‑age gaps.
Among Alzheimer’s patients, a bigger gap was associated with elevated levels of p‑tau217, a protein linked to disease pathology. Socio‑economic risk factors—including financial strain, food insecurity, limited healthcare access, adverse childhood experiences and lower educational attainment—were likewise more common among those with older‑sounding speech. The pattern held consistently across all five countries despite cultural differences.
Limitations and next steps
While the findings suggest a “huge leap toward making brain‑health assessments accessible,” experts caution that voice alone cannot serve as a definitive diagnostic tool. Dr. Manisha Parulekar, co‑director of the Center for Memory Loss and Brain Health at Hackensack University Medical Center, notes that depression, exhaustion or acute stress can also make a voice sound older, limiting specificity.
The study was cross‑sectional; most participants were evaluated at a single point rather than followed over time. Consequently, researchers could not determine whether an older‑sounding voice predicts future cognitive decline or simply reflects existing impairment. Additionally, the model was trained exclusively on Spanish‑speaking adults from five Latin American nations, so language‑specific features would need adaptation before use in other linguistic contexts.
Future research plans include longitudinal monitoring of participants and testing the approach in other languages to assess whether changes in speech can forecast dementia onset. Co‑author Adolfo García, director of the Cognitive Neuroscience Center at the University of San Andrés, emphasizes the potential of speech as a low‑cost, non‑invasive alternative to expensive brain scans or blood tests traditionally used in aging‑clock models.
Until broader validation is achieved, the speech clock may function best as an early‑warning signal that prompts clinicians to conduct more comprehensive cognitive evaluations.
Mitchell Landsberg is a Senior Technology Correspondent at News Raise. He covers consumer electronics, artificial intelligence, software developments, and digital privacy trends.
