Multilingual markers of depression in remotely collected speech samples

Abstract

Background Speech contains neuromuscular, physiological, and cognitive components and so is a potential biomarker of mental disorders. Previous studies have indicated that speaking rate and pausing are associated with major depressive disorder (MDD). However, results are inclusive as many studies are small and underpowered and do not focus on clinical samples. These studies have also been unilingual and use speech collected in highly controlled settings. If speech markers are to help understand the onset and progress of MDD, we need to uncover markers that are robust to language and establish the strength of associations in real-world data. Methods We collected speech data in 585 participants with a history of MDD in the United Kingdom, Spain, and Netherlands as part of the RADAR-MDD study. Participants recorded their speech via smartphones every two weeks for 18 months. Linear mixed models were used to identify key cross-language markers of depression from a set of 28 speech features. Findings: Increased depressive symptoms were associated with speech rate, articulation rate and intensity of speech elicited from a scripted speech task. These three features had consistently stronger effect sizes than pauses. Interpretation: Participants with more severe depressive symptoms spoke more slowly and quietly, regardless of the language used. As decreases in speech rate and articulation rate had stronger effects than any of the pausing measures tested, we speculate the observed slowing of speech is most likely due to psychomotor impairments affecting articulation rates.

Zero-shot personalization of speech foundation models for depressed mood monitoring

“…Instead, we refer the interested reader to Cummins et al. 18 for the influence of a range of acoustic features on patient health questionnaires (PHQs) and to He et al. 38 for an overview of deep learning approaches for depression recognition.…”

“… 9 , 10 Typically, these take the form of mobile monitoring applications that utilize various sensors embedded in modern smartphones or wearables and have been shown to correlate with mental states. 10 , 11 , 12 Speech, as one of the biomarkers affected by different pathologies, 13 , 14 , 15 such as mood disorders 16 and depression, 17 , 18 , 19 can be used as a means to passively monitor patients. This can be done through the pervasive recording of daily life, 20 using minimalistic models deployed in edge devices, 21 monitoring telephone conversations, 22 or eliciting responses through human-computer interaction interfaces (e.g., computer games 23 ) in a naturalistic setting.…”

Longitudinal Modeling of Depression Shifts Using Speech and Language

“…For full details on the preparation and organization of the speech data, the interested reader is referred to [7,10]. In this work, we used the free-response data only to use language features alongside acoustic information.…”

“…Speech affected by depression is often described clinically as having reduced verbal activity, shorter utterances, slower speech rate, and increased pauses [4,5]. The predictive power of individual speech features has also been linked to changes in depression symptom severity; e. g. [6,7,8,9]. Most recent speech and depression research is focused on developing machine learning models that use multivariate feature spaces to detect the presence or absence of depression in speech; e. g. [10,11,12,13,14].…”