Beyond accuracy: The need for explainable AI in biomedical voice technology

Speech and voice have emerged as valuable non-invasive biomarkers for detecting and monitoring a range of medical conditions, from neurodegenerative and respiratory diseases to psychiatric and emotional disorders. Recent advancements in artificial intelligence (AI) have accelerated this trend by enabling the identification of subtle changes in vocal patterns that elude human perception. Nevertheless, the increasing reliance on high-performing deep learning models has raised critical concerns regarding interpretability, that is an essential criterion in clinical environments, where transparency and trust are of paramount importance.