Science Advances: People with Dementia Had an Older “Speech Age”
18:52 29 October, 2026An international team of researchers has developed an algorithm that estimates a person’s age based on speech characteristics. In people with dementia, the estimated age was more likely to exceed their actual age, and the size of this gap was associated with measures of brain health. The study was published in the journal Science Advances.
The study included 2,928 Spanish-speaking participants from Argentina, Chile, Colombia, Mexico, and Peru. Of these, 1,504 had no cognitive impairment, while 1,424 had Alzheimer’s disease, frontotemporal dementia, or mild cognitive impairment.
Participants completed short speech tasks: describing events in an animation, retelling stories, and naming words according to a given rule. The algorithm analyzed hundreds of acoustic and linguistic features, ranging from speech rate and pauses to word choice and semantic specificity.
Using these data, the model predicted each participant’s age. The researchers then subtracted the person’s actual age from the estimated age to calculate their “speech age.” A positive difference meant that the person’s speech characteristics appeared older than expected.
The gap was greater in patients with dementia than in people without cognitive impairment. The largest deviations were observed in forms of frontotemporal dementia that affect language. The measure was also associated with cognitive test results and MRI-based assessments of brain aging. In the available subgroups, researchers also identified associations with certain biological markers.
According to the authors, speech analysis could become an accessible addition to studies of aging and efforts to identify signs of neurodegenerative diseases. Recording speech requires no sophisticated equipment and can be done remotely.
However, the algorithm cannot yet be considered a standalone diagnostic test. The average age-estimation error was approximately nine years, and the data were collected at a single point in time. Therefore, the study does not show whether the method can predict future dementia. Its applicability to other languages also remains to be tested.