Abstract
BACKGROUND: Mental disorders have a significant impact on many areas of people's life, particularly on affective regulation; thus, there is a growing need to find disease-specific biomarkers to improve early diagnosis. Recently, machine learning technology using speech analysis proved to be a promising field that could aid mental health assessments. Furthermore, as prosodic expressions of emotions are altered in many psychiatric conditions, some studies successfully employed a speech emotion recognition model (SER) to identify mental diseases. The aim of this paper is to discuss the utilization of speech analysis in diagnosis of mental disorders, with a focus on studies using SER system to detect mental illness. METHOD: We searched PubMed, Scopus and Google Scholar for papers published from 2014 to 2024. We conducted a preliminary search, which revealed papers on the topic. Finally, 12 studies met the inclusion criteria and were included in the review. RESULTS: Findings confirmed the efficacy of speech analysis in distinguishing between patients from healthy subjects; moreover, the examined studies underlined that some mental illnesses are associated with specific voice patterns. Furthermore, results from studies employing speech emotion recognition system to detect mental disorders showed that emotions can be successfully used as an intermediary step for mental diseases detection, particularly for mood disorders. CONCLUSION: These findings support the implementing of speech signals analysis in mental health assessment: it is an accessible and non-invasive method which can provide earlier diagnosis and a higher treatment personalization.