Computer Science Theses and Dissertations
Permanent URI for this collectionhttp://hdl.handle.net/1903/2756
Browse
5 results
Search Results
Item NUMERICAL ACOUSTICS FOR PHYSICAL AND SIMULATED ENVIRONMENTS(2023) Kaneko, Shoken Eckhart; Duraiswami, Ramani; Gumerov, Nail A; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Computer modeling and numerical analysis of acoustical phenomena have important applications including manufacturing, audio technologies in immersive multimedia, and machine learning systems involving audio. The focus of the present dissertation is the exploration of numerical methods for modeling, simulating, synthesizing, estimating, processing, controlling, and analyzing acoustical phenomena in the physical world as well as its applications to the virtual world, i.e. immersive technologies for creating virtual, augmented, and extended realities.The dissertation is structured as follows. In chapter 1, I introduce some fundamentals and basic concepts of numerical acoustics and discuss existing practical problems in acoustics. In chapter 2 and chapter 3, I propose two novel techniques for three-dimensional sound field capturing end encoding for immersive audio applications, which are both based on (semi-)analytical cancellation of scattering caused by microphone arrays mounted on acoustic scatterers. In chapter 4 and chapter 5, I introduce a fast algorithm for synthesizing acoustic impulse responses in large-scale forests, and use it to predict the performance of acoustic wildlife monitoring systems based on large-scale distributed microphone arrays. In chapter 6, I propose a novel general-purpose individual-agnostic binaural localizer which supports sound source localization from arbitrary directions without a priori knowledge of the process generating the binaural signal. In chapter 7 and chapter 8, I develop frameworks for regularized active sound control, using either point- or mode-control and using either distributed or local worn loudspeaker and microphone arrays with applications including speech privacy, personal active noise control, and local crosstalk cancellation with limited noise injection into the environment. In chapter 9, chapter 10 and chapter 11, three numerical methods for evaluating integrals arising in the (fast multipole accelerated) boundary element method are introduced. In chapter 9, a recursive algorithm is developed which allows efficient analytical evaluation of singular and nearly singular layer potential integrals arising in the boundary element method using flat high-order elements for Helmholtz and Laplace equations. In chapter 10, a differential geometry-based quadrature algorithm is developed which allows accurate evaluation of singular and nearly singular layer potential integrals arising in the boundary element method using smooth manifold boundary elements with constant densities for Helmholtz and Laplace equations. In chapter 11, an algorithm for efficient exact evaluation of integrals of regular solid harmonics over high-order boundary elements with simplex geometries is developed. In chapter 12, I discuss future research directions and conclude the dissertation.Item Numerical Geometric Acoustics(2021) Potter, Samuel Francis; Cameron, Maria K; Duraiswami, Ramani; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Sound propagation in air is accurately described by a small perturbation of the ambient pressure away from a quiescent state. This is the realm of linear acoustics, where the propagation of a time-harmonic wave can be modeled using the Helmholtz equation. When the wavelength is small relative to the size of a scattering obstacle, techniques from geometric optics are applicable. Geometric methods such as raytracing are often used for computational room acoustics simulations in situations where the geometry of the built environment is sufficiently complicated. At the same time, the high-frequency approximation of the Helmholtz equation is described by two partial differential equations: the eikonal equation, whose solution gives the first arrival time of a geometric acoustics/optics wavefront as a field; and a transport equation, the solution of which describes the amplitude of that wavefield. Phenomena related to high-frequency acoustic diffraction are frequently omitted from these models because of their complexity. These phenomena can be modeled using a high-frequency diffraction theory, such as the uniform theory of diffraction. Despite their shortcomings, geometric methods for room acoustics provide a useful trade-off between realism and computational efficiency. Motivated by the limitations of geometric methods, we approach the problem of geometric acoustics using numerical methods for solving partial differential equations. Our focus is offline sound propagation in a high-frequency regime where directly solving the wave or Helmholtz equations is infeasible. To this end, we conduct a broad-based survey of semi-Lagrangian solvers for the eikonal equation, which make the local ray information of the solution explicit. We develop efficient, first-order solvers for the eikonal equation in 3D, called ordered line integral methods (OLIMs). The OLIMs provide intuition about how to design work-efficient semi-Lagrangian eikonal solvers, but their first order accuracy is not sufficient to compute the amplitude consistently. Motivated by the requirements of sound propagation simulations, we develop higher-order semi-Lagrangian eikonal solvers which we term jet marching methods (JMMs). JMMs augment the efficiency of OLIMs by additionally transporting higher-order derivative information of the eikonal in a causal fashion, which allows for high-order solution of the eikonal equation using compact stencils. We use the information made available locally by our JMMs to use paraxial raytracing to simultaneously solve the transport equation yielding the amplitude. We initially develop a JMM which handles a smoothly varying speed of sound on a regular grid in 2D. Motivated by the requirements of room acoustics applications, we develop a second-order JMM for solving the eikonal equation on a tetrahedron mesh for a constant speed of sound as a special case. As before, we use paraxial raytracing to compute the amplitude. Additionally, we compute multiple arrivals by reinitializing the eikonal equation on reflecting walls and diffracting edges. To compute these scattered fields, we devise algorithms which allow us to apply reflection and diffraction boundary conditions for the eikonal and amplitude. For the amplitude, we construct algorithms that allow us to apply the uniform theory of diffraction in a semi-Lagrangian setting efficiently.Item Real-time Audio Reverberation for Virtual Room Acoustics(2020) Shen, Justin M; Duraiswami, Ramani; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)For virtual and augmented reality applications, it is desirable to render audio sources in the space the user is in, in real-time without sacrificing the perceptual quality of the sound. One aspect of the rendering that is perceptually important for a listener is the late-reverberation, or "echo", of the sound within a room environment. A popular method of generating a plausible late reverberation in real-time is the use of Feedback Delay Network (FDN). However, its use has the drawback that it first has to be tuned (usually manually) for a particular room before the late-reverberation generated becomes perceptually accurate. In this thesis, we propose a data-driven approach to automatically generate a pre-tuned FDN for any given room described by a set of room parameters. When combined with existing method for rendering the direct path and early reflections of a sound source, we demonstrate the feasibility of being able to render audio source in real-time for interactive applications.Item Deep Neural Networks for End-to-End Optimized Speech Coding(2017) Kankanahalli, Srihari; Jacobs, David; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Modern compression algorithms are the result of years of research; industry standards such as MP3, JPEG, and G.722.1 required complex hand-engineered compression pipelines, often with much manual tuning involved on the part of the engineers who created them. Recently, deep neural networks have shown a sophisticated ability to learn directly from data, achieving incredible success over traditional hand-engineered features in many areas. Our aim is to extend these "deep learning" methods into the domain of compression. We present a novel deep neural network model and train it to optimize all the steps of a wideband speech-coding pipeline (compression, quantization, entropy coding, and decompression) end-to-end directly from raw speech data, no manual feature engineering necessary. In testing, our learned speech coder performs on par with or better than current standards at a variety of bitrates (~9kbps up to ~24kbps). It also runs in realtime on an Intel i7-4790K CPU.Item Optimal Space-Time-Frequency Design of Microphone Networks(2014) Lai, Yenming; Balan, Radu V; Applied Mathematics and Scientific Computation; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Consider a sensing system using a large number of N microphones placed in multiple dimensions to monitor a acoustic field. Using all the microphones at once is impractical because of the amount data generated. Instead, we choose a subset of D microphones to be active. Specifically, we wish to find the D set of microphones that minimizes the largest interference gain at multiple frequencies while monitoring a target of interest. A direct, combinatorial approach - testing all N choose D subsets of microphones - is impractical because of problem size. Instead, we use a convex optimization technique that induces sparsity through a l1-penalty to determine which subset of microphones to use. Our work investigates not only the optimal placement (space) of microphones but also how to process the output of each microphone (time/frequency). We explore this problem for both single and multi-frequency sources, optimizing both microphone weights and positions simultaneously. In addition, we explore this problem for random sources where the output of each of the N microphones is processed by an individual multirate filterbank. The N processed filterbank outputs are then combined to form one final signal. In this case, we fix all the analysis filters and optimize over all the synthesis filters. We show how to convert the continuous frequency problem to a discrete frequency approximation that is computationally tractable. In this random source/multirate filterbank case, we once again optimize over space-time-frequency simultaneously.