Ein effizientes DSP Netzwerk für die Echtzeitauralisierung komplexer urbaner Szenarien



Akustische Virtuelle Realität


Masterarbeit von Andrew, Henry

Virtual reality is an increasingly relevant research area, with many applications from visualising scientific data to entertainment. It is important that a user feel immersed in the virtual environment which is presented. Though many techniques have been developed to generate visual components, the auditory rendering of a scene often leaves much to be desired. There are several prevalent techniques commonly used to auralise a scene. Wave based methods, which solve the wave equations to calculate the pressure field at the receiver location, offer high accuracy but are very slow are often restricted to low frequencies. Artificial reverberation techniques are very fast but are not physically based, and only give a very simplified approximation susceptible to many unnatural artifacts and effects. The most commonly used geometrical acoustics class of techniques offers a good balance of speed and accuracy, and has seen much development. With these techniques sound propagation is modeled as rays, and different paths they may take from the source to the receiver are found. Usually, a filter or impulse response is generated from these paths, which is convolved with an anechoic recording of the source to simulate the scene. A particularly challenging problem is that of an urban landscape. In this scene, there are likely to be many sound sources, some of which may be far away or fast moving. This means there will be a very large number of sound paths from the sources to the receiver. For example, for a single source, the sound may have a direct path to the receiver, as well as many involving reflections and diffractions around buildings and objects in the scene. This introduces a large computational load to the system. Another challenge is that the Doppler effect becomes important for fast moving sources such as cars or planes, and must be included in calculations. A "Variable Delay Line" technique can be used to simulate the Doppler effect, and reduce the length of the filters needed. However, there may still be a large number of sound paths to consider from multiple sources. Computing a large number of these is impossible on most computers with the real time constraint imposed by the virtual reality application. Many of these sound paths may carry very little importance to the overall signal, so can be disregarded to reduce computation time. The problem which I will work on with this masters thesis is how to best select the most important sound paths, and then efficiently map them to a DSP network in order to auralise a scene in real time. This will involve implementing a system in C++ which takes generated sound paths from a library and from them generates a realistic sounding auditory rendering of a scene. By adding a greater emphasis to different parts of the system, for example allowing more sound paths, or longer filters, a better overall result may be achieved. Part of the thesis will also involve determining the importance of each part of the system in the overall result. A user listening test will be conducted to aid in this.