Background: Researchers are increasingly interested in better methods for assessing the pace of aging in older adults, including vocal analysis. The present study sought to determine whether paralinguistic vocal attributes improve estimates of the age and risk of mortality in older adults.
Methods: To measure vocal age, we curated interviews provided by male U.S. World War II Veterans in the Library of Congress collection. We used diarization to identify speakers and measure vocal features and matched recording data to mortality information. Veterans (N = 2 447) were randomly split into testing (n = 1 467) and validation (n = 980) subsets to generate estimations of vocal age and years of life remaining. Results were replicated to examine out-of-sample utility using Korean War Veterans (N = 352).
Results: World War II Veterans' average age was 86.08 at the time of recording and 91.28 at the time of death. Overall, 7.4% were prisoners of war, 43.3% were Army Veterans, and 29.3% were drafted. Vocal age estimates (mean absolute error = 3.255) were within 5 years of chronological age, 78.5% of the time. With chronological age held constant, older vocal age estimation was correlated with shorter life expectancy (aHR = 1.10; 95% confidence interval: 1.06-1.15; p < .001), even when adjusting for age at vocal assessment.
Conclusions: Computational analyses reduced estimation error by 71.94% (approximately 8 years) and produced vocal age estimates that were correlated with both age and predicted time until death when age was held constant. Paralinguistic analyses augment other assessments for individuals when oral patient histories are recorded.
Keywords: Machine learning; Oral history; Paralinguistic aging; Veterans; Vocal age estimation.
© The Author(s) 2023. Published by Oxford University Press on behalf of The Gerontological Society of America. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com.