Estimation of room geometry based on impulse responses
Bachelor Thesis of Maintz, Thomas
Virtual Reality methods are used to create an immersive environment where the user is able to freely interact. Besides visual perception acoustic perception is of high importance. The process of auralization makes scenes audible and aligns the acoustic and visual feedback. For this procedure simulated room impulse responses are required which represent the room's acoustical characteristics. Room impulse responses contain early reflections within the first 50 ms. Those reflections occur on bounding surfaces and appear as peaks in room impulse responses. These peaks can be approximated by the image source model which is based on the principles of geometric acoustics. An algorithm was developed which calculates the bounding surfaces and in this way estimates the room geometry. No a priori knowledge is required except for the speed of sound and the constellation of receivers. Since the identification of higher order image sources is not trivial, mainly convex room geometries are investigated. In these spaces independent of the source and the receiver positions all first order image sources are audible. Peaks of the room impulse response are converted into estimated propagation paths using methods of Euclidean Distance Matrix and Multidimensional Scaling. Based on estimated positions of the original source and image sources bounding surfaces are calculated. In order to validate the result of the estimates, a measure of the error is introduced taking into account acoustic and geometric deviations.