SynCoPation: Synthesis Coupled Sound Propagation

Transcription

1 Online Submission ID: 0 SynCoPation: Synthesis Coupled Sound Propagation Figure 1: Our coupled sound synthesis-propagation technique has been integrated in the UnityT M game engine. We plan to demonstrate the sound effects generated by our system on a variety of scenarios: (a) Cathedral, (b) Tuscany, and (c) Game scene. In the left most case, the bowl sounds are synthesized and propagated in the Cathedral; In the middle scene, the bell sounds are synthesized and propagated in outdoor scene; In the last scene sounds of barrel hitting the ground are synthesized and propagated. 1 Abstract Sounds can augment the sense of presence and immersion of users and improve their experience in virtual environments. Recent research in sound simulation and rendering has focused either on sound synthesis or on sound propagation, and many standalone algorithms have been developed for each domain. We present a novel technique for automatically generating aural content for virtual environments based on an integrated scheme that can perform sound synthesis as well as sound propagation. Our coupled approach can generate sounds from rigid-bodies based on the audio modes and radiation coefficients; and interactively propagate them through the environment to generate acoustic effects. Our integrated system allows high degree of dynamism - it can support dynamic sources, dynamic listeners, and dynamic directivity simultaneously. Furthermore, our approach can be combined with wave-based and geometric sound propagation algorithms to compute environmental effects. We have integrated our system with the Unity game engine and show the effectiveness of fully-automatic audio content creation in complex indoor and outdoor scenes Introduction Sound simulation algorithms predict the behavior of sound waves, including generation of sound waves from vibrations and propagation of sound waves in the environment. Realistic sound simulation is important in computer games to increase the level of immersion and realism. Sound augments the visual sense of the player, provides spatial cues about the environment, and can improve the overall gaming experience. At a broad level, prior research in sound simulation can be classified into two parts - synthesis and propagation. The problem of sound synthesis deals with simulating the physical processes (e.g. vibration of a sound source) involved in generation of sound. Sound propagation, on the other hand, deals with the behavior of sound waves as they are emitted by the source, interact with the environment, and reach the listener. State-of-the-art techniques for sound simulation deal with the problem of sound synthesis and sound propagation independently. Sound synthesis techniques model the generation of sound resulting from vibration analysis of the structure of the object [Zheng and James 2009; Chadwick et al. 2009; Moss et al. 2010; Zheng and James 2010]. However, in these techniques only sound propagation in free-space (empty space) is modeled and the acoustics effects generated by the environment are mostly ignored. Sound propagation techniques [Krokstad et al. 1968; Allen and Berkley 1979; Funkhouser et al. 1998; Raghuvanshi et al. 2010; Mehra et al. 2013; Yeh et al. 2013] model the interaction of sound waves in indoor and outdoor spaces, but assume pre-recorded or pre-synthesized audio clips as input. These assumptions can result in missing sound effects and generate inaccurate (or non-plausible) solutions for the underlying physical reality produced by the process of sound simulation. For example, consider the case of a kitchen bowl falling from a countertop; the change in the directivity of the bowl with the hit position and the effect of this time-varying directivity on propagated sound field in the kitchen is mostly ignored by current simulation techniques. Similarly, for a barrel rolling down in an alley, the sound consists of multiple frequencies, where each frequency has different radiation and propagation characteristic, which are mostly ignored by current sound simulation systems. Due to these limitations, artists and game audio-designers have to manually design sound effects corresponding to these different kinds of scenarios, which can be very tedious and time-consuming. In this paper, we present the first coupled synthesis and propagation system which models the entire process of sound simulation starting from the surface vibration of objects, radiation of sound waves from these surface vibrations, and interaction of resulting sound waves with the environment. Our technique models the surface vibration characteristic of an object by performing modal analysis using the finite element method. These surface vibrations are used as boundary conditions to the Helmholtz equation solver (using boundary element method) to generate outward radiating sound fields. These radiating sound fields are expressed in a compact basis using the single-point multipole expansion [Ochmann 1999]. Mathematically, this single-point multipole expansion corresponds to a single sound source placed inside the object. The sound propagation due to this source is achieved by using numerical sound simulation technique (at low frequencies) and ray-tracing (at high frequencies). We also describe techniques to accelerate ray-tracing algorithms based on path clustering and binning. Our approach performs end-to-end sound simulation from first principles and enables automatic sound effect generation for interactive applications, thereby reducing manual effort and time-spent by artists and gameaudio designers. The main contributions of our work on coupled sound synthesispropagation include:

2 Integrated technique for accurately simulating the effect of time-varying directivity. 2. High accuracy achieved by correct phase computation, and per-frequency modeling of sound vibration, radiation and propagation. 3. Interactive runtime to handle high degree of dynamism e.g. dynamic surface vibrations, dynamic sound radiation, and sound propagation for dynamic sources and listeners. We plan to integrate our technique with the Unity T M game engine and demonstrate the effect of coupled sound synthesis-propagation on a variety of indoor and outdoor scenarios as shown in Fig boundary elements. An efficient technique known as the Equivalent source method (ESM) [Fairweather 2003; Kropp and Svensson 1995; Ochmann 1999; Pavic 2006] exploits the uniqueness of the solutions to the acoustic boundary value problem. ESM expresses the solution field as a linear combination of simple radiating point sources of various orders (monopoles, dipoles, etc.) by placing these simple sources at variable locations inside the object and matching the total generated field with the boundary conditions on the object s surface, guaranteeing the correctness of solution. [James et al. 2006] use the equivalent source method to compute the radiated field generated by a vibrating object. 2.3 Sound Propagation 94 2 Related Work Sound is a pressure wave described using the Helmholtz equation for a domain Ω: In this section, we review some of the most closely related work on sound synthesis, radiation, and propagation techniques. 2.1 Sound Synthesis CORDIS-ANIMA was perhaps the first proposed system of damped springs and masses for modeling surface vibration to synthesize physically-based sounds [Florens and Cadoz 1991]. Numerical integration using a finite element formulation was later presented as a more accurate technique for modeling vibrations [Chaigne and Doutaut 1997; O Brien et al. 2001]. Instead of using numerical integration, [van den Doel and Pai 1996; van den Doel and Pai 1998] proposed to compute analytical vibrational modes of an object, leading to considerable speedups and enabling real-time sound synthesis. [van den Doel et al. 2001] introduced the first method to determine the vibration modes and their dependence on the point of impact for a given shape, based on physical measurements. Later, [O Brien et al. 2002] presented a general algorithm to determine modal modes of an arbitrarily-shaped 3D objects by discretizing them into tetrahedral volume elements. They showed that the corresponding finite element equations can be solved analytically after suitable approximations. Consequently, they were able to model arbitrarily shaped objects and simulate realistic sounds for a few objects at interactive rates [O Brien et al. 2002]. This approach requires expensive pre-computation called modal analysis. [Raghuvanshi and Lin 2006a] used a simpler system of spring-mass along with perceptually motivated acceleration techniques to recreate realistic sound effects for hundreds of objects in real time. In this paper, we use a FEM-based method to precompute the modal modes, similar to [O Brien et al. 2002]. [Ren et al. 2012] presented an interactive virtual percussion instrument system that used modal synthesis as well as numerical sound propagation for modeling a small instrument cavity. This work, despite of some obvious similarity, is actually very different from our coupled approach. Their combination of synthesis and propagation is not well-coupled or integrated, and the volume of the underlying acoustic spaces is rather small in comparison to the typical game scenes (e.g. benchmarks shown in Fig. 1). 2 p + ω2 p = 0, x Ω (1) c2 152 where p(x) is the complex-valued pressure field, ω is the angu- 153 lar frequency, c is the constant speed of sound in a homogenous medium, and is the Laplacian operator. Boundary conditions 155 are specified on the boundary of the domain Ω by either using 156 the Dirichlet boundary condition that specifies the pressure on the 157 boundary p = f(x) on Ω, the Nuemann boundary condition that specifies the velocity of the medium p(x) 158 = f(x) on Ω, n 159 or a mixed boundary condition that specifies Z C, so that Z p(x) = f(x) on Ω. We also need to specify the behavior of n 161 p at infinity, which is usually done using the Sommerfeld radiation 162 condition [Pierce et al. 1981]: lim [ p r r + i ω p] = 0 (2) c where r = x is the distance of point x from the origin. Different methods exist to solve the equation with different formulations. Numerical methods solve for p numerically either by discretizing the entire domain or the boundary. Geometric techniques model p as a set of rays and propagate these rays through the environment Wave-based Sound Propagation Wave-based or numerical sound propagation solve the acoustic wave equation using a numerical wave solvers. These methods obtain the exact behavior of a propagating sound wave in a domain. Numerical wave solvers discretize space and time to solve the wave equation. Typically techniques include finite difference time domain (FDTD) method [Yee 1966; Taflove and Hagness 2005; Sakamoto et al. 2006], finite element method [Thompson 2006], boundary element method [Cheng and Cheng 2005], pseudo-spectral time domain [Liu 1997], and domaindecomposition [Raghuvanshi et al. 2009]. Wave-based methods have high accuracy and can simulate wave-effects such as diffraction accurately at low frequencies. However, their memory and compute requirements grow as the third or fourth power of the frequency, making them impractical for interactive applications Sound Radiation Geometric Sound Propagation The Helmholtz equation is the standard way to model sound radiating from vibrating, rigid bodies. Boundary element method is a widely used method for acoustic radiation problems [Ciskowski and Brebbia 1991; von Estorff 2000] but has a major drawback in terms of high memory requirements, i.e. O(N 2 ) memory for N Geometric sound propagation techniques use the simplifying assumption that the wavelength of sound is much smaller than features in the scene. As a result, these methods are most accurate for high frequencies and must model low-frequency effects like diffraction and scattering as separate phenomena. Commonly used 2

3 techniques are based on image source methods [Allen and Berkley 1979; Borish 1984] and ray tracing [Krokstad et al. 1968; Vorländer 1989]. Recently, there has been a focus on computing realistic acoustics in real time using algorithms designed for fast simulation. These include beam tracing [Funkhouser et al. 1998], frustum tracing [Chandak et al. 2008], and ray-based algorithms [Lentz et al. 2007; Taylor et al. 2012] that compute low-order reflections. In addition, frame-to-frame coherence of the sound field can be utilized to achieve a significant speedup [Schissler et al. 2014]. Edge diffraction effects can be approximated within GA frameworks using methods based on the uniform theory of diffraction (UTD) [Kouyoumjian and Pathak 1974] or the Biot-Tolstoy- Medwin (BTM) model [Svensson et al. 1999]. These approaches have been applied to static scenes and low-oder diffraction [Tsingos et al. 2001; Antani et al. 2012a], as well as dynamic scenes with first-order [Taylor et al. 2012] and higher-order diffraction [Schissler et al. 2014]. Diffuse reflection effects caused by surface scattering have been previously modeled using the acoustic rendering equation [Siltanen et al. 2007; Antani et al. 2012b], and radiosity-based methods [Franzoni et al. 2001]. Another commonly-used technique for ray tracing called vector-based scattering uses scattering coefficients to model diffusion [Christensen and Koutsouris 2013]. 3 Our Algorithm In this section, we give a brief background on various concepts used in the paper and present our coupled synthesis-propagation algorithm. 3.1 Background can be achieved using a fast BEM solver and specifying the Neumann Boundary Condition: p = iωρv on S, (6) n where S = Ω (the boundary of the object), ρ is the fluid density, and v is the surface s normal velocity given by v = iω(n û), where n û is modal displacement in the normal direction. This boundary condition links the modal displacements with the pressure at a point. Unfortunately, BEM is not fast enough for an interactive runtime necessitating the use of fast, approximate acoustic transfer functions [James et al. 2006]. In order to approximate the acoustic transfer, we use a source simulation technique called the Equivalent Source Method. We represent a sound source using a collection of point sources (called equivalent sources) and match the pressure values on the boundary of the object Ω with the pressure on Ω calculated using BEM. The main idea here is that if we can match the strengths of the equivalent sources to match the boundary pressure, we can evaluate the pressure at any point on Ω using these equivalent sources. Equivalent sources: The uniqueness of the acoustic boundary value problem guarantees that the solution of the free-space Helmholtz equation along with the specified boundary conditions is unique inside Ω. The unique solution p(x) can be found by expressing it as a linear combination of fundamental solutions. One choice of fundamental solutions is based on equivalent sources. An Equivalent source q(x, y i ), of the Helmholtz equation subject to the Sommerfeld radiation condition x i y i is the solution field induced at any point x due to a point source located at y i, and can be expressed as: Modal Analysis: Sound is produced by small vibrations of objects. These vibrations although invisible to the naked eye are audible if the frequency of vibration lies in the range of human hearing (20 Hz - 20 khz). Modal analysis is a well-known technique for modeling such sounds in rigid-bodies. The small vibrations can be modeled using a coupled linear system of ODEs: Kd + Cḋ + M d = f, (3) where K, C, and M are the stiffness, damping, and mass matrices respectively and f represents the (external) force vector. For small damping it is possible to approximate C as a combination of mass and stiffness matrix: C = αm+βk. This facilitates the diagonalization of the above equation, which is represented as a generalized eigenvalue problem: KU = ΛMU, (4) where Λ is the diagonal eigenvalue matrix, U contains the eigenvectors of K. Solving this eigenvalue problem enables us to write Eq. 3 as system of decoupled oscillators: q + (αi + βλ) q + Λq = U T f, (5) where U projects d into the modal subspace q with d = Uq. Acoustic transfer: The pressure p(x) at any point obtained on solving Eq. (1) is called the acoustic transfer function. The acoustic transfer function gives the relation between the surface normal displacements at a surface node and sound pressure at a given field point. A common method used in acoustics to evaluate these transfer functions through the use of boundary element method (BEM) discussed before. Since we re solving Eq. (1) in the frequency domain, we have to solve the exterior scattering problem for each mode separately. This q(x, y i ) = L 1 l l=0 m= l c ilm ϕ ilm (x) = d ik ϕ ik (x), (7) L 2 k=1 where k is a generalized index for (l, m) and c ilm is its strength. These fundamental solutions (ϕ ik ) are chosen to correspond to the field due to spherical multipole sources of order L (L = 1 being a monopole, L = 2 a dipole, and so on) located at y i. Spherical multipoles are given as a product of two functions: ϕ ilm (x) = Γ lm h (2) l (k ir i)ψ lm (θ i, φ i), (8) where (r i, θ i, φ i) is the vector (x y i ) expressed in spherical coordinates, h (2) l (k ir i) is the spherical Hankel function of the second kind, k i is the wavenumber given by ω i c, ψ lm (θ i, φ i) are the complex-valued spherical harmonics functions, and Γ lm h (2) l is the normalizing factor for the spherical harmonics. The pressure at any point in Ω due to M equivalent sources located at {y i} M i=1 can be expressed as a linear combination: p(y) = M L 1 m=l i=1 l=0 m= l c ilm ϕ ilm (y). (9) We have to determine the L 2 complex coefficients c ilm for each of the M multipoles. This compact representation of the pressure p(y) makes it possible to evaluate the pressure at any point of the domain in an efficient manner. 3.2 Coupled Algorithm We now discuss our coupled synthesis-propagation algorithm. As shown in Fig. 2, we start with the modal analysis of the sounding object which gives the modal displacements, modal frequencies, 3

4 Online Submission ID: 0 Figure 2: Overview of our coupled synthesis propagation pipeline. The bowl is used an example of a modal object. The 1st stage comprises the modal analysis. The figures in red show the first two sounding modes of the bowl. We then form an offset surface around the bowl, calculate the pressure on this offset surface, place a single multipole at the center of the object, and approximate the BEM evaluated pressure. In the runtime part of the pipeline, we use the multipole to couple with a propagation system and generate the final sound at the listener. 304 and modal amplitudes. We use these mode shapes as a boundary condition to BEM to compute the pressure on an offset surface. Then we place a single equivalent source in the center of the object and approximate the pressure calculated using BEM. This gives us a vector of (complex) coefficients of the multipole strengths. At this stage (the SPME stage in the pipeline), we have computed the representation of an acoustic radiator, which serves as source for the propagation in the runtime stage of the pipeline using either a geometric or a numeric sound propagator. Our method is agnostic to the type of sound propagator, but owing to high modal frequencies generated in our benchmarks, we use a geometric sound propagation system to obtain interactive performance. The final stage of the pipeline takes the impulse response for each mode, convolves it with that mode s amplitude, and sums it to give the final signal. We describe each stage of our pipeline below: For a single-point multipole, Eq. (8) simplifies to: Given an object, we solve the displacement equation (Eq. 5) to get a discrete set of mode shapes d i, their modal frequencies ωi, and the amplitudes qi (t). The vibration s displacement vector is given by: d(t) = Uq(t) [dˆ1,..., dˆm ]q(t), (10) where M is total number of modes and q(t) <M is the vector of modal amplitude coefficients expressed as a bank of sinusoids: We then use a Single-Point Multipole Expansion to match the pressure values, p, on Ω. This is performed by fixing the position of the multipole and iteratively increasing the order of the multipole till the error is below a certain threshold. This step has to be repeated for each modal frequency with the order generally increasing with the modal frequency. Since we are using a geometric sound propagator in the runtime stage of our pipeline, using single-point multipole (per mode) makes it possible to use just one geometric propagation source for all modes. Theoretically, each multipole should be represented as a different geometric propagation source, but since all the multipoles were kept at the same position during BEM pressure evaluation on the offset surface, we can use just one geometric propagation source and use the modal frequency (ωi ) as the filter to scale the pressure at a point. This makes it possible to have an interactive runtime performance Sound synthesis to [James et al. 2006], but use a different equivalent source representation. We first compute a manifold and closed offset around the object. This defines a clear inside to place the multipole source and also serves as the boundary Ω on which BEM solves the Helmholtz Equation to obtain pressure p at the N vertices on the surface p(y) = di t qi = ai e sin(2πfi t + θi ), where fi is the modal frequency (in Hz.), di is the damping coefficient, ai is amplitude, and θi is the initial phase Sound radiation Once we have the mode shapes and modal frequencies of an object, we compute the approximate acoustic transfer for the object similar cilm ϕilm (y). (12) l=0 m= l (11) L 1 X m=l X Since no optimal strategies exist for the optimal placement of the multipole source [Ochmann 1995], we chose the center of our modal object as the source location. This is in stark contrast to [James et al. 2006; Mehra et al. 2013] who used a hierarchical source placement algorithm to minimize the residual error. We maintain the same error thresholds as them, but simplify the problem by increasing the order L iteratively and checking the pressure

5 residual r 2 < ɛ, where r = p Ac with A being an N by L 2 multipole basis matrix, and c C L2 is complex coefficient vector where p r is the contribution from a ray r in a set of rays R. We model a multipole Ψ i using rays R as : Once we match the BEM pressure on the offset surface for each mode, we place one spherical sound source for the geometric propagation for all the modes at the same position as our multipoles. M i=1 Ψ i(x) = p(x) = r R p r(x), x Ω, (16) Sound propagation Given a single-point multipole source, we can use either a wavebased or a geometric sound propagation scheme to propagate the source s radiation pattern into the environment. We describe existing techniques that can be used with our system. Wave-based Propagation: Frequency domain numerical methods like [Mehra et al. 2013] use the equivalent source method to compute the pressure field on a domain. They decompose the scene into well-separated objects and compute the per-object and the interobject transfer functions. The per-object transfer function maps the incoming sound field incident for an object A to the outgoing field as is defined as: f(φ in A ) = T AΦ out A, (13) where Φ in A is the vector of multipoles representing the incident field on an object, Φ out A is the vector of outgoing multipoles representing the scattered field, T A is the scattering matrix containing the (complex) coefficients of the outgoing multipole sources. Similarly, the inter-object transfer function for a pair of objects A and B is defined as: ga B (Φ out A ) = G B AΦ in B, (14) where Ψ i = L 1 m=l l=0 m= l c ilmϕ ilm (y) for {i} M 1. This coupling lets us calculate the pressure for a set of ray directions sampling a sphere uniformly: For a ray direction (θ, φ), traveling a distance r, its pressure is scaled by ψ(θ, φ), h (2) l (kr), and α(ω i) where α(ω i) is the energy of a ray for a modal frequency ω i. We use a geometric ray-traced based system to get the paths and their respective energies. Path Clustering: Although, using a single geometric source reduces the number of rays considerably, in order to get the acoustic phenomena right we still need a considerable number of rays ( 15000) which makes it too slow for a modal sound source even with a few sounding modes (M 20). We solve this problem by clustering the rays based on the angle between the rays and their respective time-delays. We bin the IR (Impulse Response) according to a user-specified bin size t in seconds (Figure. 3). Then, for each bin we cluster the rays based on the binning angle ϑ. The binning algorithm is shown in Algorithm 1. Sound Intensity where G B A is the interaction matrix and contains the (complex) coefficients for mapping an outgoing field from object A to object B. In general, G B A G A B. For more details on this technique, refer to [Mehra et al. 2013] Sound Intensity Delay time The single-point multipole source can be used to represent the incident field on an object for each modal frequency, which can then be approximated using the incoming multipoles Φ in A and used in Eq. 7 to get the per-object and the inter-object transfer functions. 0 δt 2δt 3δt Delay time (binned) Figure 3: Path Clustering Geometric Propagation: These methods make the assumption that the wavelength of sound is much greater than the size of features in the scene and then treat sound as rays, frustums, or beams. Wave effects like diffraction are modeled separately using geometric approximations. We make use of the ray-based sound propagation system of [Schissler et al. 2014] to compute paths that sound can travel through the scene. This system combines path tracing with a cache of diffuse sound paths to reduce the number of rays required for an interactive simulation. The approach begins by tracing a small number (e.g. 500) of rays uniformly in all directions from each sound source. These rays strike surfaces and are reflected recursively up to a specified maximum reflection depth (e.g. 50). The reflected rays are computed using vector-based scattering [Christensen and Koutsouris 2013], where the resulting rays are a linear combination of the specularly reflected rays and random Lambertian-distributed rays. The listener is modeled as a sphere the same size as a human head. At each ray-triangle intersection, the visibility of the listener sphere is sampled by tracing a few additional rays towards the listener. If some fraction of the rays are not occuluded, a path to the listener is produced. A path contains the following output data: The total distance the ray traveled r, along with the attenuation factor α due to reflection and diffraction interactions. Diffracted sound is computed separately using the UTD diffraction model [Tsingos et al. 2001]. Given the output of the geometric propagation system, we can evaluate the sound pressure as: Algorithm 1 PathBinning(t, ϑ) 1: maxnumberofbins ceil(ir.length()/t) 2: bins.setsize(maxnumberofbins) 3: for Each ray r R do Ray directions are normalized 4: S r r.direction() 5: binindex floor(r.delay()/t) 6: bin bins[binindex] 7: for Each cluster in bin do Cluster directions are normalized 8: S c cluster.direction() check if the angle between the two vectors is less than the cluster angle ϑ 9: if S c S r > cos(ϑ) then 10: cluster.add(r) 11: end if 12: end for If the ray wasn t compatible with any of the clusters, create a new one and add the path to it 13: newcluster cluster( S r) 14: bin.add(newcluster) 15: newcluster.add(r) 16: end for 403 p(x) = r R p r(x), (15) Auralization: The last stage of the pipeline is computing the listener response for all the modes. We compute this response by 5

6 convolving the time-domain impulse response of each mode with that mode s amplitude. The final signal O(t) is: M O(t) = q i(t) IR i(t) (17) i=1 Where IR i is the impulse response of the ith mode, q i(t) is the amplitude of the ith mode, and is the convolution operator. 4 Implementation In this section, we describe the implementation details of our system. All the runtime code was written in C++ and timed on a 16- core, Intel Xeon 3.1 GHz desktop with 64 GB of RAM running Windows 7 64-bit. In the preprocessing stage, the offset surface generation and eigen decomposition code was written in C++, while the single-point multipole expansion was written in MATLAB. Preprocessing: We used finite element technique to compute the stiffness matrix K which takes the tetrahedralized model, Young s modulus and the Poisson s ratio of the sounding object and compute the stiffness matrix for the object. We then do the eigenvalue decomposition of the system using Intel s MKL library (DSYEV) and calculate the modal displacements, frequencies, and amplitudes in C++. The code to find the multipole strengths was written in MATLAB, the pressure on the offset surface was calculated using a fast BEM solver (FastBEM) using FMM-BEM (Fast multipole method). Sound Propagation: We use a fast, state-of-the-art geometric ray tracer [Schissler et al. 2014] to get the paths for our pressure computation. This technique is capable of handling very high orders of diffuse and specular reflections (e.g. 10 orders of specular reflections and 50 orders of diffuse reflections) and still maintain interactive performance. As mentioned in previous section, we cluster the rays in order to reduce the number of rays in the scene, but even with that, the pressure (i.e., the spherical harmonics and the hankel functions) computation for each ray has to be optimized heavily to meet the interactive performance requirements. Spherical Harmonic computation: The number of spherical harmonics computed per ray varies as O(L 2 ), making naive evaluation too slow for an interactive runtime. We used a modified version of available fast spherical harmonic code [Sloan 2013] to compute the pressure contribution of each ray. The available code computes only the real spherical harmonics by making extensive use of SSE (Streaming SIMD Extension). We find the complex spherical harmonics from the real ones following a simple observation: Y m l = 1 (Re(Yl m ) + ι Re(Y m l )) m > 0, (18) 2 Y m l = 1 (Re(Yl m ) ι Re(Y m l ))( 1) m m < 0. (19) 2 Using this optimized code gives us a 2-3 orders of magnitude speedup compared to existing spherical harmonic implementations, e.g., BOOST did in previous section, based on the distance traveled by them in the environment. Given a user-defined bin size δ d and the length of the IR t in seconds, we cluster ray distances into N Hankel = t δ d bins requiring us to make an order of magnitude less computations. The Hankel functions are evaluated using BOOST. Parallel computation of Mode pressure: Since each mode is independent of the other, the pressure computation for each one of them can be done in parallel. The lower modes generally require lesser time to evaluate than the higher ones, so we use a simple, scene dependent, load-balancing scheme to divide the work equally amongst all the 16 cores. We used OpenMP for the parallelizing on a multicore system. Real-Time Auralization: The final audio for the simulations is rendered using a streaming partitioned convolution technique. All audio rendering is performed at a sampling rate of 48 khz. We first construct an impulse response (IR) for each mode using the computed pressure for the paths returned by the propagation system that incorporate the effects of the single-point multipole expansion. The IR is initialized to zero and the pressure for each path is added to the IR at the sample index corresponding to the delay for that path. Once constructed, the IRs for all modes are passed to the convolution system for auralization, where they are converted to frequency domain. During audio rendering, the time-domain input audio for each mode is converted to frequency domain, then multiplied with the corresponding IR partition coefficients. The inverse FFT of the resulting sound is computed and accumulated using overlap-add in a circular output buffer. The audio device reads from the circular buffer at the current position and plays back the rendered sound. 5 Conclusion We present the first coupled sound synthesis-propagation algorithm that can generate realistic sound effects for computer games and virtual reality. We describe an approach that integrates prior methods for modal sound synthesis, sound radiation, and sound propagation. The radiating sound fields are represented in a compact basis using a single-point multiple expansion. We perform sound propagation using this source basis using fast ray-tracing to compute the impulse response and convolve them with the modes to generate the final sound at the listener. The resulting system has been integrated and we highlight the performance of many indoor and outdoor scenes. Overall, this is the first system that successfully combines these methods and can handle a high degree of dynamism in term of source radiation and propagation in complex scenes. Our approach has some limitations. It is limited to rigid objects and modal sounds. Moreover the time complexity tends to increase with the mode frequency. Our single-point multipole expansion approach can result in very high order of multipoles. The geometric sound propagation algorithm may not be able to compute the low frequency effects (e.g. diffraction) accurately in all environments. Moreover, the wave-based sound propagation algorithm involves high pre computation overhead and is limited to static scenes. Currently, we do not perform any sort of mode compression resulting in a lot of closely spaced modes being generated. We could use the compression algorithms [Raghuvanshi and Lin 2006b; Langlois et al. 2014] as a means to reduce the number of modes and thus reduce the overhead of pressure computation. Our preprocessing stage takes long time, where most of the time is spent in doing the eigen-decomposition of the stiffness matrix Distance Clustering: Even after the significant speedup achieved in calculating spherical harmonics, Hankel functions need to be computed for each ray, varying linearly with the order of the multipole. We solve this problem by clustering the paths, similar to what we There are many avenues for future work. In addition to overcoming these limitations, we would like to use it in more complex indoor and outdoor environments and generate other sound effects for complex objects in large environments (e.g. a bell ringing over a 6

7 large, outdoor valley). We would like to explore some approximate solutions to accelerate the freespace acoustic transfer computation. It would be useful to include directional sources and also accelerate the computations using iterative algorithms like Arnoldi [ARP ]. References ALLEN, J. B., AND BERKLEY, D. A Image method for efficiently simulating small-room acoustics. The Journal of the Acoustical Society of America 65, 4 (April), ANTANI, L., CHANDAK, A., TAYLOR, M., AND MANOCHA, D Efficient finite-edge diffraction using conservative from-region visibility. Applied Acoustics 73, ANTANI, L., CHANDAK, A., SAVIOJA, L., AND MANOCHA, D Interactive sound propagation using compact acoustic transfer operators. ACM Trans. Graph. 31, 1 (Feb.), 7:1 7: LENTZ, T., SCHRÖDER, D., VORLÄNDER, M., AND ASSENMACHER, I Virtual reality system with integrated sound field simulation and reproduction. EURASIP Journal on Advances in Singal Processing 2007 (January), LIU, Q. H The PSTD algorithm: A time-domain method combining the pseudospectral technique and perfectly matched layers. The Journal of the Acoustical Society of America 101, 5, MEHRA, R., RAGHUVANSHI, N., ANTANI, L., CHANDAK, A., CURTIS, S., AND MANOCHA, D Wave-based sound propagation in large open scenes using an equivalent source formulation. ACM Trans. Graph. (Apr.). MOSS, W., YEH, H., HONG, J.-M., LIN, M. C., AND MANOCHA, D Sounding liquids: Automatic sound synthesis from fluid simulation. ACM Trans. Graph. 29, 3, O BRIEN, J. F., COOK, P. R., AND ESSL, G Synthesizing sounds from physically based motion. In SIGGRAPH 01: Proceedings of the 28th annual conference on Computer graphics and interactive techniques, ACM Press, New York, NY, USA, ARPACK. BORISH, J Extension to the image model to arbitrary polyhedra. The Journal of the Acoustical Society of America 75, 6 (June), CHADWICK, J. N., AN, S. S., AND JAMES, D. L Harmonic shells: a practical nonlinear sound model for near-rigid thin shells. In ACM SIGGRAPH Asia 2009 papers, ACM, New York, NY, USA, SIGGRAPH Asia 09, 119:1 119:10. CHAIGNE, A., AND DOUTAUT, V Numerical simulations of xylophones. i. time domain modeling of the vibrating bars. J. Acoust. Soc. Am. 101, 1, CHANDAK, A., LAUTERBACH, C., TAYLOR, M., REN, Z., AND MANOCHA, D Ad-frustum: Adaptive frustum tracing for interactive sound propagation. IEEE Trans. Visualization and Computer Graphics 14, 6, CHENG, A., AND CHENG, D Heritage and early history of the boundary element method. Engineering Analysis with Boundary Elements 29, 3 (Mar.), CHRISTENSEN, C., AND KOUTSOURIS, G Odeon manual, chapter 6. CISKOWSKI, R. D., AND BREBBIA, C. A Boundary element methods in acoustics. Computational Mechanics Publications Southampton, Boston. FAIRWEATHER, G The method of fundamental solutions for scattering and radiation problems. Engineering Analysis with Boundary Elements 27, 7 (July), O BRIEN, J. F., SHEN, C., AND GATCHALIAN, C. M Synthesizing sounds from rigid-body simulations. In The ACM SIGGRAPH 2002 Symposium on Computer Animation, ACM Press, OCHMANN, M The source simulation technique for acoustic radiation problems. Acustica 81, OCHMANN, M The full-field equations for acoustic radiation and scattering. The Journal of the Acoustical Society of America 105, 5, PAVIC, G A technique for the computation of sound radiation by vibrating bodies using multipole substitute sources. Acta Acustica united with Acustica 92, (15). PIERCE, A. D., ET AL Acoustics: an introduction to its physical principles and applications. McGraw-Hill New York. RAGHUVANSHI, N., AND LIN, M. C Interactive sound synthesis for large scale environments. In SI3D 06: Proceedings of the 2006 symposium on Interactive 3D graphics and games, ACM Press, New York, NY, USA, RAGHUVANSHI, N., AND LIN, M. C Interactive sound synthesis for large scale environments. In Proceedings of the 2006 symposium on Interactive 3D graphics and games, ACM, RAGHUVANSHI, N., NARAIN, R., AND LIN, M. C Efficient and accurate sound propagation using adaptive rectangular decomposition. Visualization and Computer Graphics, IEEE Transactions on 15, 5, fastbem making efficient high-fidelity acoustic modeling a reality! fastbem.com/fastbemacoustics.html. FLORENS, J. L., AND CADOZ, C The physical model: modeling and simulating the instrumental universe. In Represenations of Musical Signals, G. D. Poli, A. Piccialli, and C. Roads, Eds. MIT Press, Cambridge, MA, USA, FRANZONI, L. P., BLISS, D. B., AND ROUSE, J. W An acoustic boundary element method based on energy and intensity variables for prediction of highfrequency broadband sound fields. The Journal of the Acoustical Society of America 110, RAGHUVANSHI, N., SNYDER, J., MEHRA, R., LIN, M. C., AND GOVINDARAJU, N. K Precomputed Wave Simulation for Real-Time Sound Propagation of Dynamic Sources in Complex Scenes. SIGGRAPH , 3 (July). REN, Z., MEHRA, R., COPOSKY, J., AND LIN, M. C Tabletop ensemble: touch-enabled virtual percussion instruments. In Proceedings of the ACM SIG- GRAPH Symposium on Interactive 3D Graphics and Games, ACM, SAKAMOTO, S., USHIYAMA, A., AND NAGATOMO, H Numerical analysis of sound propagation in rooms using the finite difference time domain method. The Journal of the Acoustical Society of America 120, 5, FUNKHOUSER, T., CARLBOM, I., ELKO, G., PINGALI, G., SONDHI, M., AND WEST, J A beam tracing approach to acoustic modeling for interactive virtual environments. In Proc. of ACM SIGGRAPH, JAMES, D. L., BARBIČ, J., AND PAI, D. K Precomputed acoustic transfer: output-sensitive, accurate sound generation for geometrically complex vibration sources. In ACM SIGGRAPH 2006 Papers, ACM, New York, NY, USA, SIG- GRAPH 06, KOUYOUMJIAN, R. G., AND PATHAK, P. H A uniform geometrical theory of diffraction for an edge in a perfectly conducting surface. Proceedings of the IEEE 62, 11, KROKSTAD, A., STROM, S., AND SORSDAL, S Calculating the acoustical room response by the use of a ray tracing technique. Journal of Sound and Vibration 8, 1 (July), KROPP, W., AND SVENSSON, P. U Application of the time domain formulation of the method of equivalent sources to radiation and scattering problems. Acta Acustica united with Acustica 81, 6, LANGLOIS, T. R., AN, S. S., JIN, K. K., AND JAMES, D. L Eigenmode compression for modal sound models. ACM Transactions on Graphics (TOG) 33, 4, SCHISSLER, C., MEHRA, R., AND DINESH, M High-order diffraction and diffuse reflections for interactive sound propagation in large environments. In Proc. of ACM SIGGRAPH. SILTANEN, S., LOKKI, T., KIMINKI, S., AND SAVIOJA, L The room acoustic rendering equation. The Journal of the Acoustical Society of America 122, 3 (September), SLOAN, P.-P Efficient spherical harmonic evaluation. Journal of Computer Graphics Techniques (JCGT) 2, 2 (September), SVENSSON, U. P., FRED, R. I., AND VANDERKOOY, J An analytic secondary source model of edge diffraction impulse responses. Acoustical Society of America Journal 106 (Nov.), TAFLOVE, A., AND HAGNESS, S. C Computational Electrodynamics: The Finite-Difference Time-Domain Method, Third Edition, 3 ed. Artech House Publishers, June. TAYLOR, M., CHANDAK, A., MO, Q., LAUTERBACH, C., SCHISSLER, C., AND MANOCHA, D Guided multiview ray tracing for fast auralization. IEEE Transactions on Visualization and Computer Graphics 18, THOMPSON, L. L A review of finite-element methods for time-harmonic acoustics. The Journal of the Acoustical Society of America 119, 3,

8 TSINGOS, N., FUNKHOUSER, T., NGAN, A., AND CARLBOM, I Modeling acoustics in virtual environments using the uniform theory of diffraction. In Proc. of ACM SIGGRAPH, VAN DEN DOEL, K., AND PAI, D. K Synthesis of shape dependent sounds with physical modeling. In Proceedings of the International Conference on Auditory Displays. VAN DEN DOEL, K., AND PAI, D. K The sounds of physical shapes. Presence 7, 4, VAN DEN DOEL, K., KRY, P. G., AND PAI, D. K Foleyautomatic: physicallybased sound effects for interactive simulation and animation. In SIGGRAPH 01: Proceedings of the 28th annual conference on Computer graphics and interactive techniques, ACM Press, New York, NY, USA, VON ESTORFF, O Boundary elements in acoustics: advances and applications, vol. 9. Wit Pr/Computational Mechanics. VORLÄNDER, M Simulation of the transient and steady-state sound propagation in rooms using a new combined ray-tracing/image-source algorithm. The Journal of the Acoustical Society of America 86, 1, YEE, K Numerical solution of initial boundary value problems involving maxwell s equations in isotropic media. IEEE Transactions on Antennas and Propagation 14, 3 (May), YEH, H., MEHRA, R., REN, Z., ANTANI, L., MANOCHA, D., AND LIN, M Wave-ray coupling for interactive sound propagation in large complex scenes. ACM Trans. Graph. 32, 6, 165:1 165:11. ZHENG, C., AND JAMES, D. L Harmonic fluids. ACM Trans. Graph. 28, 3, ZHENG, C., AND JAMES, D. L Rigid-body fracture sound with precomputed soundbanks. In SIGGRAPH 10: ACM SIGGRAPH 2010 papers, ACM, New York, NY, USA,