New AI Tool Analyzes Speech to Detect Dementia Risk and Biological Aging

An experimental artificial intelligence tool analyzes hundreds of speech and language patterns to estimate a person’s chronological age and flag potential dementia risks, according to a study published Sept. 30 in the journal Science Advances and reported by livescience.com. Researchers found that adults whose estimated speech age exceeded their actual chronological age faced higher rates of cognitive impairment, signs of accelerated biological aging, and less favorable socioeconomic conditions.

Building the Speech-Age Gap Model

Scientists used data from the ReD-Lat consortium to evaluate more than 2,900 Spanish-speaking adults ranging in age from 18 to 88 across Argentina, Chile, Colombia, Mexico, and Peru. About 1,500 participants were cognitively healthy, while the remainder exhibited mild cognitive impairment, Alzheimer’s disease, or frontotemporal dementia. Each participant completed seven verbal tasks, including describing an animated video, rapid word-retrieval exercises, and story recall tests.

Researchers recorded, transcribed, and analyzed more than 700 distinct features from the audio files, encompassing pauses, pitch, speaking speed, vocabulary choice, and emotional expression. A machine-learning model was trained on these features alongside the participants’ actual ages. By comparing the AI-generated age estimates against true ages, researchers calculated a metric they termed the “speech-age gap.”

Correlating Voice Patterns With Biomarkers

Cognitively healthy participants displayed the smallest disparities between their chronological age and the AI-predicted age. Larger gaps appeared in individuals with mild cognitive impairment and dementia, with the most pronounced discrepancies occurring in patients with language-dominant frontotemporal dementia. Participants with older-sounding speech also scored lower on attention, language, and memory tests, and struggled more with daily functioning.

Blood tests revealed parallel biological indicators. Larger speech-age gaps tracked with accelerated aging on three epigenetic clocks, which measure chemical tags on DNA. Among Alzheimer’s patients, widened gaps also correlated with elevated levels of p-tau217, a protein marker associated with the disease. Wider gaps tracked with dementia risk factors such as limited education, financial strain, food insecurity, and difficult childhoods across all five countries.

Evaluating Clinical Limitations and Uncertainty

While external specialists view the technology as a promising avenue for broad neurological screening, they emphasize that the system cannot serve as a standalone diagnostic tool.

Dr. Manisha Parulekar, Hackensack University Medical Center

Dr. Manisha Parulekar, co-director of the Center for Memory Loss and Brain Health at Hackensack University Medical Center in New Jersey, noted that the model remains restricted by its training data. Because many subjects were assessed at a single point in time, researchers could not establish whether an older voice predicts future cognitive decline or merely tracks disease progression. In addition, the current algorithm functions exclusively in Spanish and requires extensive adaptation before it can be applied to English or other language groups. The research team aims to adapt the speech-processing algorithms for other languages and conduct longitudinal studies to track patients over extended periods.

More on this story: AI Model Detects Type 2 Diabetes Risk from a 20-Second Voice Sample · New Three-Minute Brain Test Could Revolutionize Alzheimer's Early Detection