A method for predicting droplet emission from surrounding A-weighted sound pressure level during speech using a generalized linear mixed model with speech level as a mediator

While the Wells-Riley model, internationally used for infection probability prediction, treats viral emission (a generation parameter) as a constant based on human respiratory activity and activity level, there remains room for further investigation to improve the validity of infection probability assessment when applied to real-world environments. Therefore, in this study, to propose new droplet emission parameters for infection probability prediction models that consider human speech characteristics and droplet generation mechanisms, speech experiments were conducted in a clean room, where experimental participants spoke under varying A-weighted sound pressure levels of sound environment from 45 to 70 dB. Non-linear regression analysis was performed to establish the relationship between the surrounding A-weighted sound pressure level during speech and the speech level. Furthermore, to clarify the relationship between speech level and droplet emission, regression analysis was conducted with speech level as the explanatory variable and aerosol and droplet emission volumes as objective variables. Goodness-of-fit tests for aerosol volume frequency revealed that droplet emission frequency follows a Gamma-Poisson distribution. Based on these findings, a prediction method for droplet emission, mediated by speech level and driven by the A-weighted sound pressure level, was derived using a generalized linear mixed model.