Voice as a Digital Biomarker

The human voice contains rich physiological and neurological information that can serve as a powerful, non-invasive digital biomarker for disease detection. Our research focuses on developing advanced artificial intelligence methods to analyze vocal signals for scalable, low-cost, and accessible medical screening.

A key challenge in voice-based diagnostics is ensuring that AI models remain reliable across different recording environments, devices, languages, and patient populations. We address this through domain generalization strategies that enable models to maintain performance even when applied to new and unseen conditions. By improving robustness and transferability, we aim to bridge the gap between laboratory research and real-world clinical deployment.

In parallel, we investigate the discovery of novel voice biomarkers — subtle acoustic and phonatory patterns that reflect underlying physiological or neurological changes. By combining signal processing, machine learning, and clinical insight, our work advances voice as a practical tool for early detection, risk assessment, and remote health monitoring.

Our long-term vision is to establish voice-based AI systems as reliable, clinically meaningful digital health technologies that enhance accessibility and support data-driven medical decision-making.