SwasthyaVani: A Voice-Enabled Multilingual Framework for Disease Prediction and Healthcare Access in Rural India

Abstract

Access to timely healthcare remains a critical challenge in rural India, where physician shortages, infrastructure deficits, and language diversity collectively hinder effective medical intervention. This paper presents SwasthyaVani (meaning Voice of Health), a voice-enabled multilingual clinical decision support system designed specifically for rural populations in India. The system accepts free-form symptom descriptions via speech or text in Telugu, Hindi, and English, processes them through a custom multilingual natural language processing pipeline, and classifies the input across 133 disease categories using an ensemble of machine learning models. Three classifiers were rigorously evaluated: Random Forest, XGBoost, and a Deep Neural Network (DNN). The DNN achieved the highest classification accuracy of 91.97%, with macro-averaged precision of 92.1%, recall of 91.8%, F1-score of 91.9%, and ROC-AUC of 0.994. The system further integrates GPS-based geolocation for nearby healthcare facility discovery and maintains longitudinal health records per authenticated user. Ethical safeguards are embedded throughout: the system explicitly positions itself as a decision-support aid, not a diagnostic replacement. Experimental results confirm strong per-class discrimination across all 133 disease classes with minimal inter-class confusion. SwasthyaVani demonstrates that accessible, language-native AI tools can meaningfully improve health-seeking behavior in underserved rural communities.