ki:elements

Speech Biomarkers Modelling Depression Severity

Felix Menne1, Felix Dörr1, Julia Schräder2,3, Lisa Wagels2,3, Ebru Baykara1, Johannes Tröger1, Alexandra König1,4,5, Ute Habel2,3

1ki:elements GmbH, Saarbrücken, Germany; 2Department of Psychiatry, Psychotherapy and Psychosomatics, Faculty of Medicine, RWTH Aachen, Germany; 3Institute of Neuroscience and Medicine: JARA-Institute Brain Structure Function Relationship (INM 10), Research Center Jülich, Jülich, Germany; 4Cobtek (Cognition-Behaviour- Technology) Lab, University Côte d’azur, Nice, France; 5Université Côte d’Azur, Centre Hospitalier et Universitaire, Clinique Gériatrique du Cerveau et du Mouvement, Centre Mémoire de Ressources et de Recherche, Nice, France

* Poster presented at the Bridge2AI Voice Conference, USA

Background: Psychiatry faces a challenge in lacking objective biomarkers, relying on subjective assessments. Automated speech analysis shows promise in detecting affective states in depressed patients. This project aims to identify discriminating speech features between patients with major depression (MDD) and healthy controls (HC), examining correlations with symptom severity measures.

Methods: 44 MDD patients and 52 HC were recruited from the Psychiatry Department, University Hospital Aachen, Germany. Participants narrated positive and negative life events, recorded for analysis. Beck Depression Inventory (BDI-II) and Hamilton Rating Scale for Depression gauged depression severity. Transcribed audio recordings underwent feature extraction. Spearman Rank Sum Correlations with BDI-II scores and Kruskal-Wallis tests for clinical staging were computed. In order to find features robust both in construct and clinical validity, we identified those which emerged as significant in all analyses.

Results: From 124 speech features, five features consistently emerged as significant (Table 1). Only features from negative storytelling remained significant across all tests.

Conclusion: Our study identified robust acoustic features associated with MDD status and its severity. In future, these findings may shape voice-based biomarkers, enhancing clinical diagnosis and MDD severity monitoring.

Share this article