Speech analysis for detecting depression in older adults: A Systematic Review

Israel Martínez-Nicolás^1*Diego Criado¹Fernando Gordillo León¹Francisco Martinez²Juan J. G. Meilán^1,3

• ¹Faculty of Psychology, University of Salamanca, Salamanca, Spain
• ²Universidad de Murcia Facultad de Psicologia y Logopedia, Murcia, Spain
• ³Instituto de Neurociencias de Castilla y Leon, Salamanca, Spain

Background: Depression is highly prevalent among older adults, exceeding rates in the general population. Traditional diagnostic tools, such as interviews and self-reports, are limited by subjectivity, time demands, and overlap with age-related changes. Speech, as a non-invasive behavioral marker, is promising for objective depression assessment, but its specific utility in older populations remains less explored. This systematic review identifies speech characteristics linked to depression in older adults and their clinical potential. Methods: Following PRISMA guidelines, a search was conducted in Medline, CINAHL, PsychINFO, IEEE, and Web of Science for studies published in the last ten years. Eligible studies included adults aged over 55, with depression diagnosis or symptoms, and at least one acoustic variable. Sixteen studies met inclusion criteria. Methodological quality was assessed with JBI tools, and speech parameters and classification outcomes were extracted. Results: Depressed older adults consistently showed slower speech rate, longer and more variable pauses, reduced intensity, and altered voice quality. Predictive studies using machine learning reached accuracies of 76%-95%, particularly when age and gender were controlled. Findings were inconsistent for F0 and formants: women often showed lower peak frequency and amplitude, while men displayed higher amplitude change and formant frequencies. Limitations included small clinical samples and insufficient control of confounders, especially cognitive impairment. Conclusion: Speech analysis appears reliable, non-invasive, and cost-effective for detecting depression in older adults. Temporal, prosodic, and spectral features show strong diagnostic potential. Further research with larger, representative samples is required to validate speech-based biomarkers as complements to existing assessments.