Assessing Chronic Stress, Coping Skills, and Mood Disorders through Speech Analysis: A Self-Assessment 'Voice App' for Laptops, Tablets, and Smartphones

Background: Computerized speech analysis (CSA) is a powerful method that allows one to assess stress-induced mood disturbances and affective disorders through repeated measurements of speaking behavior and voice sound characteristics. Over the past decades CSA has been successfully used in the clinical context to monitor the transition from 'affectively disturbed' to 'normal' among psychiatric patients under treatment. This project, by contrast, aimed to extend the CSA method in such a way that the transition from 'normal' to 'affected' can be detected among subjects of the general population through 10-20 self-assessments. Methods: Central to the project was a normative speech study of 5 major languages (English, French, German, Italian, and Spanish). Each language comprised 120 subjects stratified according to gender, age, and education with repeated assessments at 14-day intervals (total n = 697). In a first step, we developed a multivariate model to assess affective state and stress-induced bodily reactions through speaking behavior and voice sound characteristics. Secondly, we determined language-, gender-, and age-specific thresholds that draw a line between 'natural fluctuations' and 'significant changes'. Thirdly, we implemented the model along with the underlying methods and normative data in a self-assessment 'voice app' for laptops, tablets, and smartphones. Finally, a longitudinal self-assessment study of 36 subjects was carried out over 14 days to test the performance of the CSA method in home environments. Results: The data showed that speaking behavior and voice sound characteristics can be quantified in a reproducible and language-independent way. Gender and age explained 15-35% of the observed variance, whereas the educational level had a relatively small effect in the range of 1-3%. The self-assessment 'voice app' was realized in modular form so that additional languages can simply be 'plugged in' once the respective normative data become available