The tool examines hundreds of speech and language patterns to estimate a person’s age based on how they speak. In a study of nearly 3,000 Spanish-speaking adults, people whose estimated “language age” was older than their actual age were more likely to have cognitive problems, dementia and signs of accelerated biological aging.

The findings, reported Wednesday, September 30, in Science Advances, also linked larger language-age gaps with less favorable social and economic circumstances.

How does an AI voice clock work?

The speech analysis system is similar to other biological aging clocks, which use biological markers to estimate a person’s risk of age-related disease.

Researchers used data from the ReD-Lat Consortium, a large dementia research project in Latin America. The study included more than 2,900 Spanish-speaking adults from Argentina, Chile, Colombia, Mexico and Peru. Participants were between 18 and 88 years old.

About 1,500 participants were cognitively healthy. The remaining participants had mild cognitive impairment, Alzheimer’s disease or frontotemporal dementia.

Each participant completed seven tasks, including describing an animated video, naming as many words, vegetables and animals as possible in 60 seconds, and retelling a short story immediately and again 20 to 30 minutes later.

Scientists recorded and transcribed the participants’ speech, then identified more than 700 characteristics. These included pauses, speaking rate, pitch, vocabulary and emotional expression.

The researchers combined these speech features with each participant’s chronological age to train a machine-learning model. The model estimated the person’s age from their speech. Researchers then compared the predicted age with the participant’s actual age to calculate a “language age gap.”

Older-sounding speech was linked to cognitive problems

The AI model found that some participants sounded older than expected for their chronological age. Cognitively healthy people had the smallest differences between their actual age and their estimated language age.

Participants with mild cognitive impairment or dementia had larger gaps. The biggest differences appeared among people with language-dominant frontotemporal dementia, a condition that primarily affects communication.

People with larger language-age gaps also performed worse on tests of memory, language and attention. They were more likely to have difficulty with everyday activities.

Blood tests showed a similar pattern. For three measures of an epigenetic clock—which estimates biological age by examining chemical tags attached to DNA—larger language-age gaps were associated with faster biological aging.

Among participants with Alzheimer’s disease, larger gaps were also associated with higher levels of p-tau217, a protein linked to the disease.

The language-age gap was also correlated with several dementia risk factors, including economic hardship, food insecurity, limited access to health care, difficult childhood experiences and less education. The pattern was similar across all five countries despite social and cultural differences, according to study co-author Adolfo Garcia, director of the Center for Cognitive Neuroscience at the University of San Andres in Argentina.

“Existing aging clocks are very powerful, but many rely on brain scans and blood tests that are expensive, invasive or difficult to repeat regularly. All these limitations can be overcome by speech.”

Adolfo Garcia, study co-author

What the AI voice clock cannot tell us yet

The findings do not mean that an older-sounding voice can diagnose dementia. The human voice is sensitive to many aspects of overall health, including depression, exhaustion and severe stress.

Dr. Manisha Parulekar, co-director of the Center for Memory and Brain Health at Hackensack University Medical Center in New Jersey, was not involved in the study. She said audio measurements might eventually help prompt doctors to examine a patient’s cognitive health more closely, but should not be used alone to diagnose dementia.

“The human voice is very sensitive to our overall health. In other words, a person’s speech may simply sound ‘older’ because they are extremely depressed, exhausted or under severe life stress.”

Manisha Parulekar, Center for Memory and Brain Health, Hackensack University Medical Center

Many participants were assessed at only one point in time rather than being followed for years. As a result, the researchers could not determine whether an older language age predicts future cognitive decline or whether specific speech changes develop as dementia progresses.

The model was also trained and tested exclusively on Spanish-speaking participants from five Latin American countries. It will need to be adapted and tested in other languages and populations before it can be used more broadly.

“You can’t just take the Spanish model and incorporate English into it,” Parulekar said. Language-specific features need to be adapted and tested in other groups.

The researchers plan to study the approach in additional languages and follow participants over time. Their goal is to determine whether changes in speech can indicate future cognitive decline.

“I truly believe that there are very strong signals in audio that predict future dementia and conversion,” Garcia said. “But we need more and more diverse research.”

This article is for informational purposes only and does not provide medical advice.

Hernandez, H., Pedraza, L.K., Santamaria García, H., Mogilner, S., Regas, A., Prado, P., Cuadros, J., Amoruso, L., González, L., Delavere, D., Espinoza-Puels, J.P., Espinoza, J.P., Jahn, C., Matussi, F., Caccia, M., Welford, A.S., Perera, N., Inshauspe, J.F., Ferrante, F.J., … Ibanez, A. (2026). “Audio clocks decipher dementia phenotypes, social exposomes, and biological aging.” Science Advances, 12(40), eaef9864.