Network Flow Modeling of Speech-Induced Facemask Leakage

Facemasks are a cornerstone of respiratory disease mitigation in high-risk settings such as hospitals, airports, and schools, yet most evaluations overlook a critical factor: the impact of speech. Talking introduces complex facial movements that alter mask fit, airflow leakage, and jet dynamics, but these effects remain poorly understood. We address this gap with a fast, machine learning enhanced flow dynamics framework that combines facial morphology modeling, phoneme-driven deformations, physics-based mask deployment, and fluid dynamics simulations to predict leakage and jet velocities during speech. The facemask interface is represented as a network of interconnected porous channels with compatibility constraints, incorporating deformations from 60 phoneme-driven words. Our results show velocity amplification of 1.5×–5.6× relative to baseline breathing and enable the creation of a "speech aerodynamics dictionary" linking word characteristics to leakage patterns. These insights advance understanding of how speech affects mask performance and can inform better mask design and public health strategies for real-world use.