Sex-Based Speech Pattern Recognition for Post-traumatic Stress Disorder

Post-Traumatic Stress Disorder (PTSD) is a mental health disorder diagnosed by physical and psychological evaluations, scales of assessment, and diagnosis criteria. PTSD diagnosis can be affected by subjective evaluations and limited accessibility. Hence, alternative methodologies for diagnosing PTSD have emerged, like speech analysis, it is recognized as a source of predictors to psychiatric disorders, can be evaluated remotely and suggest sex-based differences. Few studies have been focused on changes in the speech of PTSD population, and, as far as we can ascertain do not explored features of sex-based speech associated with PTSD. For this reason, the present study aims to automatically detect sex-based speech patterns linked to indicators of PTSD through the characterization of speech signals. This research employed decision trees to recognize speech predictors of PTSD. The resulting models achieved favorable performance, being the women’s model the most accurate. Moreover, when not considering sex-based differences, the spectral domain features highlighted and confirmed that PTSD population is more likely to have uniform speech. Additionally, the results suggest sex-based differences. Such as, the men’s model considers the dispersion of speech and if a voice is tense, and the women’s model focuses on the frequency bands of the speech.