Depression and Anxiety Screening for Pregnant Women via Free Conversational Speech in Naturalistic Condition

Depression and anxiety during pregnancy are critical public health concerns, impacting both maternal and fetal well-being. The onset of these mental health conditions during pregnancy can lead to adverse outcomes, such as preterm birth and impaired infant development. Traditional diagnostic and monitoring methods, reliant on self-reporting and clinical evaluations, are often hindered by stigmatization, subjective bias, and accessibility issues, leading to underdiagnosis and inadequate management of these conditions. This study addresses the challenge by making use of free conversational speech recordings via smartphones as an approach for screening depression and anxiety in pregnant women. By leveraging machine learning techniques to analyze speech patterns, this research seeks to offer a non-invasive, objective, and accessible method for early detection and intervention. The research involved collecting conversational speech samples from pregnant women attending a high-risk pregnancy outpatient service, employing smartphones to capture naturalistic speech in everyday contexts. Machine learning algorithms were utilized combined with different audio feature sets to analyze these recordings in conjunction with PHQ-4 questionnaire scores. The combination of Mel Frequency Cepstral Coefficients and OpenSmile feature sets analyzed through LightGBM classifiers emerged as the most effective, achieving an accuracy of 75% and an Area Under the Curve of 0.71. These findings highlight the potential of smartphone-based speech analysis as a viable screening tool for mental health in pregnant women, suggesting a significant shift towards accessible, real-time monitoring. This technological intervention could enhance maternal health by enabling early detection and timely care, ultimately improving pregnancy outcomes and child development.