Four setups are demonstrated (columns) for each FNO trained (rows). For the multi-source domains this shows time-step t = 40, and for all others this shows t = 72. Multi-source experiments contain between 1 and 4 simultaneous sound sources. Sound source positions are marked in cyan. Scattering and B+S experiments contain a reflective object of a different size. Boundary experiments feature one randomly positioned sound source. Ground-truth and predicted values are converted to decibels with a reference signal strength of unity. (a) Ground-truth FDTD simulations produced using SRL and IWB stencils. (b) FNO predictions: each prediction from an FNO was obtained by providing the early time-steps of simulations pictured in (a) as input data. (c) Absolute difference in decibels between FNO predictions and ground-truth FDTD simulations.