| Acoustic phenomena have a large impact on our everyday lives, from influencing our enjoyment of music in a concert hall, to affecting our concentration at school or work, to potentially negatively impacting our health through deafening noises. The sound that reaches our ears is absorbed, reflected, and filtered by the shape, topology, and materials present in the environment. However, many computer simulation techniques for solving these sound propagation problems are either computationally expensive or inaccurate. Additionally, the costs of some methods are dramatically increased in design optimization processes in which several iterations of sound propagation evaluation are necessary.;The primary goal of this dissertation is to present techniques for efficiently solving the sound propagation problem and related optimization problems for computer-aided design. First, we propose a parallel method for solving large acoustic propagation problems, scalable to tens of thousands of cores. Second, we present two novel techniques for optimizing certain acoustic characteristics such as reverberation time or sound clarity using wave-based sound propagation. Finally, we show how hybrid sound propagation algorithms can be used to improve the performance of acoustic optimization problems and present two algorithms for noise minimization and speech intelligibility improvement that use this hybrid approach.;All the algorithms we present are evaluated on various benchmarks that are computer models of architectural scenes. These benchmarks include challenging environments for existing sound propagation algorithms, such as large indoor or outdoor scenes, structural complex scenes, or the prevalence of difficult-to-model sound propagation phenomena. Using the techniques put forth in this dissertation, we can solve many challenging sound propagation and optimization problems on the scenes in an efficient manner. We are able to accurately model sound propagation using wave-based approaches up to SI{10}{kilohertz} (the full range of human speech) and for the full range of human hearing (22kHz) using our hybrid approach. Our noise minimization methods show improvements of up to 13dB in noise reduction on some scenes, and we show a 71% improvement in speech intelligibility using our algorithm. |
