Multimodal Acoustic-Linguistic Machine Learning for Postpartum Depression Screening Using the Edinburgh Postnatal
James Griffin1, Bretton H Talbot1, Loren Larsen1
1Videra Health, Inc., Orem, Utah, USA.
Women'S Health Reports (New Rochelle, N.Y.)
|August 1, 2026
Summary
This study developed a speech-based model to efficiently predict postpartum depression (PPD) scores using acoustic and linguistic features. The findings show promise for early PPD risk identification and clinical workflow enhancement.
Area of Science:
- Speech analysis
- Computational linguistics
- Mental health screening
Background:
- Postpartum depression (PPD) is a significant mood disorder affecting women after childbirth.
- PPD is often underdiagnosed and undertreated despite available interventions.
- Accurate and early detection of PPD is crucial for timely clinical intervention.
Purpose of the Study:
- To develop an efficient method for predicting Edinburgh Postnatal Depression Scale (EPDS) scores.
- To utilize combined acoustic and linguistic features from speech for PPD risk assessment.
- To create a tool that can augment clinical workflows for PPD screening.
Main Methods:
- Recruited 275 pregnant and postpartum women aged 18+.
- Collected audio responses via a smartphone app for EPDS score prediction.
- Employed a multimodal approach combining linguistic and acoustic speech features.
Main Results:
- Achieved 84.4% sensitivity and 76.9% specificity for PPD screening at the ≥10 threshold.
- Demonstrated a strong correlation (R=0.749) between predicted and actual EPDS scores.
- Reported an area under the curve of 0.886, indicating robust model performance.
Conclusions:
- The developed model shows strong performance in identifying PPD risk from speech.
- This technology can enhance clinical workflows by prioritizing at-risk individuals.
- Speech-based PPD screening offers a novel and accessible approach to mental health assessment.