Scientists build a 'speech clock' that estimates ageing from voice
Scientists have built a 'speech clock': a machine-learning model that predicts how well a person is ageing from characteristics of their voice and how they talk. The findings were published in the journal Science Advances and reported by Nature's news team.
To build it, neuroscientist Agustín Ibáñez of Adolfo Ibáñez University in Santiago, a co-author of the study, and his colleagues recorded 2,928 Spanish speakers from Argentina, Chile, Colombia, Mexico and Peru while they completed various speech tasks. The group was a mixture of healthy people and people with mild cognitive impairment, Alzheimer's disease or other forms of dementia. Machine-learning algorithms extracted more than 700 speech characteristics from the audio, such as pitch and vocabulary range, features that can change with ageing and dementia. The researchers then used these data to train the model to predict each participant's age. The clock draws on hundreds of vocal features, including pitch and talking speed.
Using the clock, the team calculated a 'speech age gap' for each person: the difference between the age the clock predicts and the person's chronological age. Healthy people were generally assessed as having speech that matched their chronological age. People with cognitive issues were categorised as sounding older than expected, and large speech age gaps were strongly associated with cognitive issues such as those that arise in dementia. Overall, the clock could distinguish healthy individuals from those with some form of cognitive impairment. The article says this suggests the clock could be a useful tool for determining whether a person is growing older faster than expected.
The idea is new in one respect. Existing ageing clocks usually rest on biological markers: 'brain clocks' use neuroimaging signatures to judge whether a brain is ageing faster than its owner's years suggest, and 'epigenetic clocks' read patterns of methyl tags on DNA to estimate biological age. According to the article, researchers had not so far developed a clock based on speech. Ibáñez says speech is a promising window because speaking involves a "huge amount of brain work", and that the team sees "a huge predictive value, just with a very simple four minutes of speech recordings".
Jed Meltzer, a cognitive neuroscientist who specializes in language at the University of Toronto in Canada and was not involved in the study, says the clock could be a boon for tracking ageing in low-resource regions, because it does not rely on expensive or invasive technologies such as brain scans and blood tests. He called it "a very impressive piece of work."
Key facts
- The 'speech clock' is a machine-learning model that predicts a person's age from their voice and speech, and was published in Science Advances.
- It was built from recordings of 2,928 Spanish speakers from Argentina, Chile, Colombia, Mexico and Peru, healthy and with cognitive impairment or dementia.
- Algorithms extracted more than 700 speech characteristics, such as pitch and vocabulary range, to train the model.
- Large gaps between speech-predicted age and chronological age were strongly associated with cognitive issues such as those in dementia.
- Ibáñez cites four minutes of speech as enough for a huge predictive value; outside expert Jed Meltzer sees promise for low-resource regions.
Why it matters
Ageing clocks so far have leaned on biological markers: brain clocks built from neuroimaging and epigenetic clocks built from DNA methylation patterns. According to the article, no one had yet built one from speech. Speaking involves a 'huge amount of brain work', in Ibáñez's words, so voice may reveal how the brain is ageing without a scanner or a blood draw.
Who it affects
The study covers Spanish speakers from five Latin American countries, including healthy people and people with mild cognitive impairment, Alzheimer's disease or other dementias. Meltzer says the approach could especially help track ageing in low-resource regions, where brain scans and blood tests are expensive or hard to get.
How to use it
This is a research result, not a product. The source names no tool for the public and gives no timescale for clinical or commercial use. What it does describe is the input: Ibáñez says a very simple four minutes of speech recordings already carries huge predictive value, and the recordings in the study came from various speech tasks.
How solid is it
The work appeared in the peer-reviewed journal Science Advances and rests on 2,928 recorded speakers. Meltzer, who was not involved, calls it a very impressive piece of work. The source gives no numeric accuracy figures, such as error in years or correlation values, and the original article is paywalled past its opening section.
Risks and caveats
The finding is an association. Large speech age gaps were strongly associated with cognitive issues, which suggests the clock could be useful; the source does not say it can diagnose dementia. It also does not say whether the clock was validated on non-Spanish speakers or other populations. Because the article is cut off by a paywall, any further limitations discussed later in it are not visible.
“We can see a huge predictive value, just with a very simple four minutes of speech recordings”
— Agustín Ibáñez, neuroscientist at Adolfo Ibáñez University in Santiago and co-author of the study