Acoustic Simulation in Real World Scenes

Major Goals: The objective of this project is to develop new approaches for acoustic modeling and rendering in reconstructed models of real-world scenes. These include water-tight reconstruction of large or non-line-of-sight features that govern the propagation of sound. We will also develop better methods to automatically classify the acoustic properties of different materials in terms of frequency-dependent absorption, scattering, and diffusion coefficients. In order to faithfully reproduce the acoustic effects in real-world scenes, we also need faster methods to simulate late reverberation effects as well diffuse reflections and diffraction. Our proposed research borrows ideas from computer vision, machine learning, signal processing, computer graphics, psycho-acoustics and scientific computing to address these challenges with respect to acoustic modeling and rendering. Furthermore, the use of sound sensors and reconstruction techniques can also be used to improve the performance of computer vision algorithms. Accomplishments: Accomplishments: We have developed novel algorithms:1. Fast sound propagation algorithms that combine ray racing and reverberation filters for dynamic scenes. 2. New hybrid sound propagation and acoustic optimization algorithms that are used to place the speakers so that the real-world audio effects are clear. 3. New methods based on machine learning for estimating the direction of arrival estimation of sound in the real-world scenarios. 4. New methods for dynamic sound field synthesis in virtual and real world scenes using acoustic optimization. 5. New methods based on machine learning to approximate acoustic scattering fields and use them for interactive sound propagation.