AI “Voice Clock” Predicts Biological Aging and Detects Cognitive Decline

Scientists have developed a “voice clock” that can estimate a person’s age from the characteristics of their voice and the way they speak. The tool analyzes hundreds of audio features, including pitch and speaking rate, to predict a person’s age.

Researchers used the clock to calculate a person’s “vocal age gap” — the difference between their age predicted by the voice clock and their actual age. A large vocal age gap was strongly associated with cognitive problems, including those that occur in dementia. The findings suggest that the tool could help identify people who are aging faster than expected.

The results were published today in the journal Science Advances¹.

“We see that a very simple four-minute audio recording can have a large predictive value,” says study co-author Agustín Ibáñez, a neuroscientist at Adolfo Ibáñez University in Santiago.

Jed Meltzer, a cognitive neuroscientist specializing in language at the University of Toronto in Canada who was not involved in the study, says the voice clock could help track aging in low-resource areas because it does not require expensive, invasive techniques such as brain scans or blood tests. “It’s a very impressive work,” he says.

How the voice clock predicts age

Aging clocks use biological markers to assess how quickly a person’s body is changing. For example, a brain clock uses neuroimaging signatures to determine whether an individual’s brain is aging faster than their chronological age would suggest. An epigenetic clock analyzes methyl tags on DNA to estimate biological age.

Until now, researchers had not developed an aging clock based on audio. Ibáñez says a voice-based tool could provide new insight into the aging process because speaking involves “an enormous amount of mental labor.”

Study used nearly 3,000 Spanish speakers

To create the voice clock, Ibáñez and his colleagues recorded 2,928 Spanish speakers from Argentina, Chile, Colombia, Mexico and Peru as they completed various vocal tasks. The participants included healthy people as well as people with mild cognitive impairment, Alzheimer’s disease or other forms of dementia.

The researchers used machine-learning algorithms to extract more than 700 speech features that can change with age and dementia. These included characteristics such as pitch and vocabulary range. The data were then used to train an audio model to predict each participant’s age.

Voice age was linked to cognitive impairment

Overall, the voice clock distinguished between healthy participants and people with some forms of cognitive impairment. The speech of people with cognitive problems was classified as older than expected based on their chronological age, whereas healthy participants were generally rated as having speech consistent with their actual age.