The voice may carry physiological information related to age. The Voice Age study made available by npj Aging on September 11, 2026 used 30-second recordings to build age-prediction models in 6,979 Hebrew-speaking Israeli adults aged 40–70, and examined links between voice age and other physiological features.
A prediction error of roughly four years
The researchers built separate models for women and men, with mean absolute errors in chronological-age prediction of 3.95 and 4.41 years respectively. Age-adjusted deviations in voice age were also associated with some measures of body composition, sleep-related breathing and other physiology. The paper is currently an accepted, citable advance-publication version. Abstract and version information
Mean absolute error describes how far predicted values are from actual ages. The roughly four years here is the model’s age-prediction error in the study, not an error interval for an individual’s disease risk.
Voice age was only partly correlated with other age clocks. The authors reported that adding voice information to models based on other types of data contributed additional age-prediction information. This suggests different measurements may capture different aspects of the body.
Easy sampling is an advantage that still needs validation
A 30-second recording places relatively little burden on participants. Moving from one language and population to everyday monitoring, however, requires validation across languages, devices and recording conditions. Whether age prediction can help assess health changes requires long-term outcome studies.
In TASCAP’s view, the next value of this work depends on two things: whether the voice signal can be reproduced in independent populations, and how much useful information it adds beyond existing measures. The study offers leads for voice biomarkers; it has not turned a recording into an individual health diagnosis.