Frequency- and Time-dependent Geometry for Real-time Auralizations

The quality of present-day room acoustic simulations stands and falls by the quality of the underlying CAD room models. A ‘high-quality’ room model does not implicate that it has to be highly detailed with a lot of small objects and ornamentation. High accuracy of a model and its auralization is only achieved when basic acoustic principles are regarded. This means for geometrically based simulations that wavelengths from 1.7cm till 17m have to be handled with regard to their reflection/scattering pattern at room or object surfaces. As this spans the dimensions of the majority of known objects, walls etc., only an adapting room model that changes its level of detail accordingly to the incident sound wave frequency can ensure correct results, e.g. for low-frequency specular reflections. When it comes to real-time auralizations, which are used in sophisticated virtual reality systems, another emerging aspect is how to deal with very complex room geometries in limited computation time. Using active frequency dependent geometry has great advantages in this discipline due to much faster simulations when simple geometries with a low polygon count are involved. Lower frequencies typically travel much longer than higher ones in common rooms and thus they produce the majority of computation load which is now significantly reduced, yielding a great speed-up potential. Going on, the introduction of a temporal discretization reducing room model details step by step over the duration of the room impulse response can save valuable computation resources as well. This technique makes use of perceptual characteristics of the human ear, which is not able to distinguish fine structures in the late part of the impulse response. Furthermore, most of the energy in this late part originates from diffuse reflections for which the exact geometry does not matter. Thus, the active geometry model switches to simpler structures for late reflections. In this contribution, the newly developed active geometry model, which uses a frequency and time dependent level of detail, will be presented as well as results from comparative listening tests. These could point out the necessary complexity for the highest detail step as well as the maximally allowed simplification based on human perception.