Photo Credit: Visions

The following is a summary of “Voice of depression: speech features as biomarkers for major depressive disorder,” published in the November 2024 issue of Psychiatry by Menne et al. 

Psychiatry lacks objective biomarkers, relying on subjective assessments. Automated speech analysis shows potential for detecting symptom severity in depression. 

Researchers conducted a retrospective study to identify speech features differentiating patients with major depressive disorder (MDD) from healthy controls (HCs) and their association with symptom severity. 

They recruited 44 patients with MDD and 52 HCs. Depression severity was assessed using BDI-II and the Hamilton Rating Scale. Audio recordings were analyzed for acoustics, speech rate, and content. Machine learning models combined speech features and neuropsychological assessments to differentiate patients with MDD from HCs. 

The results showed significant differences in pitch and loudness between patients with MDD and HCs (effect sizes 2 = 0.183–0.3, P < 0.001). Temporality, lexical richness, and speech sentiment also showed moderate to high effect sizes (2 = 0.062–0.143, P < 0.02). A support vector machine (SVM) model with 10 acoustic features achieved high performance (AUC = 0.93), like the BDI-II model (AUC = 0.99, P = 0.01). 

They identified speech features linked to MDD, with a machine learning model performing similarly to traditional assessments. These findings may enhance clinical diagnosis and monitoring of MDD. 

Source: bmcpsychiatry.biomedcentral.com/articles/10.1186/s12888-024-06253-6