Overview
- The study, published Wednesday in Science Advances, trained machine-learning models on voice recordings from 2,928 Spanish speakers across Argentina, Chile, Colombia, Mexico and Peru to build a ‘speech clock’ that estimates a speaker’s age.
- Researchers calculated a speech age gap—the difference between voice-predicted age and actual age—and found larger gaps were linked to worse cognition and a higher likelihood of dementia diagnoses.
- The speech age gap correlated with independent biological measures including MRI-derived brain aging, three DNA-methylation epigenetic clocks, and higher plasma p-tau217 levels in people with Alzheimer’s disease.
- The models used more than 700 acoustic and linguistic features such as pitch, pause length, speech rate, vocabulary richness and semantic precision, and also reflected lifelong social adversity like low education and food insecurity.
- Authors and reporters stress the evidence is cross-sectional only and say longitudinal, cross-linguistic, real-world validation plus work on bias, privacy and clinical pathways are required before the method can be used in practice.