Just noticeable difference for simulation accuracy between full and reduced order models (L).

Area of Science:

• Computational acoustics
• Psychoacoustics
• Numerical methods

Background:

• Model order reduction (MOR) techniques accelerate computational acoustics simulations by simplifying complex wave phenomena.
• A trade-off exists between simulation speed and physical accuracy in MOR, but the perceptual impact of these simplifications is not well understood.

Purpose of the Study:

• To determine the physical error limit in reduced-order room acoustic models that remains perceptually indistinguishable to human listeners.
• To establish a threshold for acceptable error in computationally accelerated acoustic simulations.

Main Methods:

• Simulations of various two-dimensional room geometries and reverberation times were conducted using reduced-order models.
• A three-alternative forced-choice listening test was employed to assess human perception of simulation accuracy.
• Statistical analysis was used to identify significant perceptual differences between full and reduced-order model stimuli.

Main Results:

• The perceptual threshold for error in reduced-order room acoustic models was found to be between 1% and 0.1% relative root mean square error.
• Stimuli generated with a relative root mean square error below this threshold did not result in statistically significant perceptual differences.
• Above this threshold, human listeners could reliably distinguish between the original and reduced-order model simulations.

Conclusions:

• MOR techniques can be employed for significant computational acceleration in room acoustics.
• A specific error range (0.1%–1% RMSE) exists where MOR approximations are perceptually acceptable.
• Understanding this perceptual limit is crucial for developing efficient yet accurate room acoustic simulation tools.