Overview

• The exploratory Frontiers in Digital Health study used 12,523 recordings from the public Bridge2AI-Voice dataset to analyze acoustic features in 306 participants.
• Harmonic-to-noise ratio variability emerged as the most informative marker, distinguishing cisgender men with healthy voices, benign lesions and early laryngeal cancer by pitch and noise measures.
• Analysis failed to produce clear diagnostic patterns for women and could not reliably separate benign or malignant lesions from other vocal disorders.
• Researchers stressed that the findings are preliminary and require larger, professionally labeled, gender-balanced datasets to improve statistical power.
• Authors estimate that with expanded data and clinical validation, AI-driven voice screening tools could enter pilot testing within a few years.