Speech age may reveal more about how a person is ageing than the sound of their voice alone suggests. A new study has found that the difference between a person’s actual age and the age estimated from their speech is associated with changes in cognition, brain health, biological ageing and dementia.
Researchers developed a computer-based “speech clock” that uses machine learning to estimate a person’s chronological age from the way they speak. The system examines hundreds of features in both speech and language, including speaking speed, pauses, pitch, emotional expression and vocabulary. The researchers then compare the estimated age with the person’s real age to calculate what they call a “speech age gap”.
A small gap suggests that a person’s speech appears broadly consistent with their chronological age. A larger positive gap means the system has judged their speech to appear older than expected. The measure does not mean that someone is literally ageing by a certain number of extra years. Instead, it is a statistical difference between an age predicted by a computer model and a person’s actual age.
The study, published in Science Advances, analysed recordings from 2,928 Spanish-speaking adults in Argentina, Chile, Colombia, Mexico and Peru. The participants included cognitively healthy adults as well as people with mild cognitive impairment, Alzheimer’s disease and different forms of frontotemporal dementia.
The researchers asked participants to complete several speaking tasks. These included describing an animated film, recalling a story and producing words from particular categories. The recordings gave the researchers enough information to examine both how participants spoke and the language they used.
More than 700 acoustic and linguistic characteristics were considered by the machine-learning models. Some captured the mechanics of speech, such as pitch, pace and pauses. Others reflected language use, including vocabulary and the precision and organisation of what participants said.
The resulting speech age gap showed a clear relationship with cognitive health. People without cognitive impairment generally had the smallest differences between their predicted speech age and chronological age. Larger gaps were found among people with cognitive impairment, Alzheimer’s disease and frontotemporal dementia, with the largest differences appearing in the language-dominant form of frontotemporal dementia.
The relationship was not limited to language tests. Participants with older-appearing speech also tended to perform worse on measures of memory, attention and other cognitive abilities. This is important because it suggests that the speech signal may reflect broader changes in brain function rather than simply difficulty with speaking.
The researchers also compared speech age with measures obtained through other forms of medical and biological assessment. Larger speech age gaps were associated with brain ageing measured through neuroimaging and with several biological ageing measures. Among participants with Alzheimer’s disease, an older speech profile was also associated with higher levels of p-tau217, a blood biomarker linked to Alzheimer’s pathology.
Agustin Ibanez, a senior author of the study and a professor of brain health associated with Trinity College Dublin and the Global Brain Health Institute, said the findings suggest that speech contains information about several aspects of ageing at the same time.
“Our voice appears to contain much more information about aging than we previously recognised,” Ibanez said. He said speech can reflect the effects of chronological ageing as well as changes involving cognition, the brain, broader biology and a person’s accumulated social environment.
That combination is what makes the speech clock potentially useful. Conventional assessments of brain and biological ageing can involve MRI scans, blood tests or specialist clinical examinations. A speech-based assessment, by comparison, could potentially be carried out using an ordinary recording device and analysed remotely by software.
Researchers and outside experts say this could eventually be useful in places where access to advanced medical equipment is limited. Cognitive neuroscientist Jed Meltzer of the University of Toronto, who was not involved in the research, told Nature that the approach could be particularly valuable for tracking ageing in low-resource settings because it does not depend on expensive or invasive technologies.
But the findings do not mean that an older-sounding voice can currently be used to diagnose dementia or determine whether a particular person will develop the disease.
The study is largely a snapshot of the relationship between speech and other measures of ageing and disease. It showed that people with certain conditions tended to have larger speech age gaps, but it did not establish that an older speech profile can predict who will later develop dementia. Researchers will need longer-term studies to determine whether changes in speech can identify cognitive decline before symptoms become apparent.
There are also questions about how well the system would work outside the population in which it was developed. The study focused on Spanish speakers from five Latin American countries, so further research will be needed to test speech clocks in other languages, cultures and populations. Differences in accents, vocabulary, education, social background and communication styles could affect the way a computer interprets speech.
The researchers’ findings nevertheless point to an increasingly important idea in ageing research: information about the brain may not always require a brain scan to detect. Everyday abilities such as speaking draw on several processes at once, including memory, attention, language, planning and the organisation of information.
For now, the speech clock remains a research tool rather than a medical test. Its value lies in the association it has revealed between speech age and a range of brain, cognitive and biological measures. Whether a short recording of someone’s speech can eventually help doctors identify early cognitive changes, monitor ageing or complement existing dementia assessments will require further testing.
The study does, however, suggest that the way people speak may carry a measurable record of changes occurring elsewhere in the body and brain. What sounds like an ordinary conversation could contain information that current technology is only beginning to learn how to read.
