[Automated audio analysis and depression: A systematic umbrella review].

Background and purpose: The early and accurate diagnosis of depression is essential for the timely initiation of appropriate treatment. However, the use of traditional diagnostic methods is often subjective, labour-intensive, and time-consuming. Automated voice analysis offers promising solution to overcome these challenges by enabling the objective measurement of voice-based biometric features. The aim of this study is to review the research practice and findings of voice analysis methods supported by machine learning, to uncover inconsistencies, and to formulate constructive suggestions for future research.

Methods: This paper presents the results of a so-called umbrella review, which integrates the findings of already published systematic literature reviews and meta-analyses. For the identification of publications, we used the PubMed, Scopus, and ProQuest databases, following the PRISMA guidelines. The search interval covered the last 5 years. The search was conducted using predefined search terms and selection criteria, with the involvement of independent reviewers. Prior to detailed analysis, the methodological quality of the publications was assessed using the AMSTAR2 evaluation system.

Results: Through the systematic literature search, we identified a total of 162 unique records. Based on the inclusion and exclusion criteria, 6 publications were selected for detailed analysis. The results highlight the background factors limiting the applicability of the developed models and also emphasize the importance of acoustic characteristics that can be identified as biomarkers of depression despite methodological inconsistencies. This review supports the importance of machine learning and voice analysis in advancing the diagnostics of depression. However, to translate research outcomes into practice, beyond the application of standardized methods, validation across diverse test groups is necessary, among other things.

Conclusion: The application of machine learning in depression detection promises numerous advantages, such as objective diagnosis or early detection. This technology could offer cost-effective solution in the long run while providing greater access to mental health services. Nevertheless, the field is still evolving, and further research is needed to enhance the reliability of these methods.