Abstract:
The current state of the art of virtual reality allows a user to interact with computer-generated visual worlds and provides a high level of realism. The immersion can be increased significantly by additionally simulating the corresponding room acoustics. A correct acoustical impression is reached by the reproduction of binaural signals and already satisfies high demands at the present, but for interactive systems it is mostly limited to sources located in a free-field, i.e. the filters used for computation are applicable only for anechoic environments.

Current real-time capable room acoustical simulations for virtual environments are mostly based on the principles of geometrical acoustics. However, these approaches do not provide an exact reproduction of the room's sound properties under real-time conditions due to high computation demands and are therefore limited to the simulation of just plausible sound fields. Hence, faster algorithms are required to further enhance the room acoustical simulation.

This work presents a modification of the commonly known image source method, which reduces the computation demands of this highly time-expensive approach remarkably. Binary space partitioning trees are used for a fast determination of audible image sources, and an efficient data structure with respect to the real-time auralization process is introduced. Thus, a faster determination of the binaural room impulse response is gained, which is essential for a real-time capable auralization process. In addition, geometrical relations are used more efficiently to prevent the generation of redundant image sources and to increase the performance of this algorithm. Furthermore, experiments have been carried out to show the current hardware limitations for this modified image source method.