Metamodel predictions for long range speech intelligibility using parabolic equation calculations

The Crank–Nicholson parabolic equation (CNPE) method was used to calculate speech intelligibility for long range propagation of voice messages. How intelligible a voice message will be at long range is very sensitive to atmospheric conditions and parameters such as source height and background noise levels. In order to identify this sensitivity, four independent variables were identified and varied throughout the study. These independent variables are the atmospheric refraction index, which takes into account both temperature and wind gradients within the atmosphere, ambient temperature, source height, and background noise level. CNPE computations were performed for 31 optimally chosen cases where all four of the independent variables were random variables and were inputted into a metamodel. This allows for curve fitting and extrapolation of the existing data to determine intermediate results for any operation scenario defined within the bounds of the current study. Upon doing this, it was determined that the atmospheric refraction index is the most sensitive variable. In addition, there is virtually no dependence on ambient surface temperature.