An AI predicting room acoustics is like a student taking an exam with a cheat sheet. If the test points are the same as the training ones, the model simply memorizes the acoustic 'fingerprint' of each position — like a student recognizing a question by the dog-eared corner. Tighten the rules — group the data by location and hide frequency clues — and accuracy plummets from 85% to 9–57%.
After honest validation, the algorithms seem equally helpless. But AI retains superiority in measuring loudness and reveals an unexpected talent: reconstructing the sound field between microphones with up to 88% accuracy. The paradox is that the harsh test failures highlighted its true strength — real-time interpolation. By the way, the world's longest echo lasts 112 seconds (in an oil tank in Scotland); such an algorithm could model it from just a few measurements.
In concert halls, sound decay time rarely exceeds 2 seconds, otherwise music turns into cacophony.