Lumped-Element Model Predicts Subglottal Pressure from High-Speed Laryngeal Video

The model treats the respiratory system as interconnected piston-cylinder units with nonlinear viscoelastic lung tissue and flow resistances. High-speed videoendoscopy from one normophonic subject supplied the glottal area waveform via a convolutional network, driving vocal-fold oscillation and aerodynamic coupling. Resulting simulations produced time-varying subglottal pressures and energy transfers that cannot be measured noninvasively in humans. The approach extends classic lumped-parameter respiratory models by embedding AI-extracted kinematics, enabling patient-specific extensions for phonation disorders.

Existing vocal-fold models often rely on simplified Bernoulli flow or two-mass approximations that ignore upstream lung compliance. This framework closes that gap by propagating compressible airway dynamics into the glottis, revealing how lung viscoelasticity modulates pressure pulses. It therefore supplies mechanistic links between respiratory mechanics and voice production that purely acoustic or endoscopic studies miss.

Future clinical translation hinges on multi-subject validation against direct tracheal pressure recordings. Larger cohorts with voice pathologies will test whether the model distinguishes healthy from disordered phonation at thresholds useful for diagnosis. Regulatory acceptance will require prospective trials showing that simulated pressures correlate with treatment outcomes at least as well as current invasive or empirical methods.

The work highlights an emerging pattern: AI-derived boundary conditions are replacing population averages in physiological simulation, moving computational voice science toward individualized prediction.