Capturing Voices at Risk : Designing Semi-Structured Speech Tasks for Voice Data Collection in Depression and Anxiety Risk Groups

Background : The human voice carries rich multimodal information, disclosing cognitive and emotional states through linguistic content and acoustic properties. Voice data has emerged as a promising digital biomarker for detecting anxiety and depression in at-risk populations. However, collecting high-quality voice data that effectively captures vocal manifestations of psychological distress remains a significant methodological challenge.<br/>Methods : We developed “MindCheck,” a conversation-based semi-structured speech task designed specifically for at-risk depression and anxiety populations. Our two-phase approach first collected user requirements through interviews to establish user experience (UX) desiderata for voice data acquisition, then designed and validated a human subject task to capture informationally rich voice data from individuals at risk.<br/>Results : Iterative task refinement improved user engagement and reduced task completion difficulties. Participants emphasized the importance of engaging interactions and clear guidance, which resulted in more adjustable interface configurations. Higher engagement levels and more natural interaction patterns suggest improved circumstances for capturing authentic voice biomarkers.<br/>Conclusions : This study highlights the critical role of user-centered design in developing effective voice data collection protocols for anxiety and depression risk assessment. The iterative task refinement approach demonstrated that prioritizing user engagement creates better circumstances for authentic voice biomarker capture, enhancing the potential of conversation-based tools for early detection in vulnerable populations.