1. (The University of Tokyo) (Hokkaido University)
Real-time Rendering of Aerodynamic Sound Using Sound Textures based on Computational Fluid Dynamics
Tsuyoshi Yamamoto
Tomoyuki Nishita
(The University of Tokyo)
Yoshinori Dobashi
(Hokkaido University)
北海道大学の土橋です｡
雲および大気の散乱光を考慮した稲妻の効率的なレンダリング法について発表させていただきます。

2. Examples of aerodynamic sound
Hokkaido University
Sound of wind
Sound generated by swinging objects quickly

3. Overview Introduction Related Work
Principle and Prediction of Aerodynamic Sound
Basic Idea of Our Method
Computation of Sound Texture
Real-time Sound Rendering
Examples
Conclusions

4. Overview Introduction Related Work
Principle and Prediction of Aerodynamic Sound
Basic Idea of Our Method
Computation of Sound Texture
Real-time Sound Rendering
Examples
Conclusions

5. Introduction Simulation of virtual environments
Sound: important element
voice, contact sound, etc.
Improving reality of virtual environments
Use of recorded sound
need to find suitable sound
quality depends on environment

6. Introduction Physically-based sound synthesis
compute waves based on physical simulation
generate sound automatically according to object motion
Limited to sound due to solid objects
Sound due to fluid
wind(aerodynamic sound), water, explosion ...

7. Real-time rendering of aerodynamic sound
Goal and Feature
Real-time rendering of aerodynamic sound
source is not oscillation of solid objects
creating sound textures for aerodynamic sound
rendering sound in real-time according to object motion

8. Real-time sound rendering
Goal and Feature
Sound by swinging sword and club
Real-time sound rendering
Sound synthesis depending on shapes and motion of objects

9. Overview Introduction Related Work
Principle and Prediction of Aerodynamic Sound
Basic Idea of Our Method
Computation of Sound Texture
Real-time Sound Rendering
Examples
Conclusions

10. No methods for aerodynamic sound
Related Work in CG
Propagation of sound
[Takala92] [Funkhouser99] [Tsingos01]
simulate reflection/absorption due to objects to compute sound taking into account geometric relation between source and receiver
[Hahn95][O’Brien01] [O’Brien02] [van den doel01]
Synthesis of sound waves
compute sound waves by numerical analysis of subtle oscillation of objects
No methods for aerodynamic sound

11. Related Work in CFD Prediction of aerodynamic sound Our method…
[Lele97]
to reduce noise due to high-speed transportation facilities, etc.
complex numerical fluid simulation
not appropriate for real-time applications
Our method…
Makes use of methods developed in CFD
Realizes real-time sound synthesis

12. Overview Introduction Related Work
Principle and Prediction of Aerodynamic Sound
Basic Idea of Our Method
Computation of Sound Texture
Real-time Sound Rendering
Examples
Conclusions

13. Principle and Prediction
Source of aerodynamic sound
cylinder　
flow　
vortices in air
vortices
subtle fluctuations of air pressure due to vortices
Prediction method
Lighthill’s basic theory in 1952
[Ligh52]
numerical simulation of compressible Navier-Stokes equations
→ computationally expensive
Curle’s model

14. Curle’s Model Prediction by behavior of air near objectpa
sound
sound source field
flow　
vortex　
object　
receiver q　 r
center position o　
Incompressible fluid analysis

15. Curle’s Model
Prediction by behavior of air near object

16. Curle’s Model Prediction by behavior of air near object normal
sound source function
(SSF)
time
amp.
flow
sound source field
normal
pressure
g(t) (x component)

17. Curle’s Model Prediction by behavior of air near object pa g(t)
sound source function
(SSF) pa
sound pressure
receiver q
sound source field
flow r
g(t)
center position o

18. Curle’s Model Prediction by behavior of air near object
sound source function
(SSF)
constraint: Size of object must be sufficiently small relative to wavelength of sound

19. Overview Introduction Related Work
Principle and Prediction of Aerodynamic Sound
Basic Idea of Our Method
Computation of Sound Texture
Real-time Sound Rendering
Examples
Conclusions

20. Basic Idea Use of Curle‘s model not applicable to large object
subdivide object into small regions
equivalent to assuming independent virtual point sound sources
receiver q
region 1
region 2
region n
virtual
sound
source +

21. Basic Idea Computing sound texture (preprocess)
Rendering aerodynamic sound (real-time)

22. sound texture: w(l, s, u, v)
Basic Idea
Computing sound texture (preprocess)
fluid analysis
→ table of sound source func.
Rendering aerodynamic sound (real-time)
s : time in texture domain
t : time in reality l
sound texture: w(l, s, u, v)
speed v
time s
sound source pos. l
SSF table
fluid analysis
uniform
flow
direction u
speed v

23. sound texture: w(l, s, u, v)
Basic Idea
Computing sound texture (preprocess)
fluid analysis
→ table of sound source func.
Rendering aerodynamic sound (real-time)
sound texture: w(l, s, u, v)
fluid analysis l
flow
direction u
speed v

24. Basic Idea Computing sound texture (preprocess)
fluid analysis
→ table of sound source func.
Rendering aerodynamic sound (real-time)
dir./speed
→sound texture
→values of SSF
move v1 v2 vn l c1 c2 cn
(sound texture)
SSF values
(g1, g2, …, gn)
receiver pos. q

25. Basic Idea Computing sound texture (preprocess)
fluid analysis
→ table of sound source func.
Rendering aerodynamic sound (real-time)
dir./speed
→sound texture
→values of SSF
→Curle’s model
→ sound pressure
SSF values
(g1, g2, …, gn)
sound wave
Curle‘s model
receiver pos. q

26. Basic Idea Computing sound texture (preprocess)
fluid analysis
→ table of sound source func.
Rendering aerodynamic sound (real-time)
dir./speed
→sound texture
→values of SSF
→Curle’s model
→ sound pressure
SSF values
(g1, g2, …, gn)
sound wave
Curle‘s model
receiver pos. q

27. Overview Introduction Related Work
Principle and Prediction of Aerodynamic Sound
Basic Idea of Our Method
Computation of Sound Texture
Real-time Sound Rendering
Examples
Conclusions

28. Computation of Sound Texture
Fluid analyses for many directions and speeds
long computation time
sound texture
speed v
time s
sound source pos. l
fluid analysis
flow
dir. u
speed v

29. Computation of Sound Texture
Properties of aerodynamic sound
frequency ∝ flow speed v
amplitude ∝ (flow speed v )6
Need only sound texture at base speed v0
sound texture
speed v
time s
sound source pos. l
fluid analysis
flow
dir. u
Reduce computation time and memory requirement drastically v0

30. Choosing 2D or 3D Fluid Analysis
Stick-like object
sound source pos.
speed v
time v0
2D analysis
cross section x y
flow

31. Choosing 2D or 3D Fluid Analysis
Stick-like object
2D analysis
cross section x y
flow
speed v
time v0
sound source pos.
1D sound tex.

32. Choosing 2D or 3D Fluid Analysis
Stick-like object 1 2
direction
time 3
point sound
source 2 3 1
1D sound tex.
2D sound tex.
2D analysis
Others
speed v
time
sound source pos. v0
3D analysis
point sound source
flow
2D sound tex.

33. Choosing 2D or 3D Fluid Analysis
Stick-like object
2D sound tex.
2D analysis
point sound
source
1D sound tex. 1 2 3
direction
time 1 2 3 1 2 3
Others
2D sound tex.
3D analysis

34. Choosing 2D or 3D Fluid Analysis
Stick-like object
2D sound tex.
2D analysis
point sound
source
1D sound tex. 1 2 3
direction
time 3
direction
time
sound source pos.
Others 2 1
3D sound tex.
2D sound tex.
3D analysis

35. Overview Introduction Related Work
Principle and Prediction of Aerodynamic Sound
Basic Idea of Our Method
Computation of Sound Texture
Real-time Sound Rendering
Examples
Conclusions

36. Real-time Sound Rendering
Procedure
- repeat for each time step Dt

37. Real-time Sound Rendering
Procedure
- repeat for each time step Dt
1. compute direction ci and speed vi
move v1 v2 vn c1 c2 cn

38. Real-time Sound Rendering
Procedure
- repeat for each time step Dt
direction
(c1, c2, …, cn)
speed
(v1, v2, …, vn)
w(l, s, u, v0)
sound texture
values of SSF
(g1, g2, …, gn)
1. compute direction ci and speed vi
2. compute SSF gi

39. Real-time Sound Rendering
Procedure
- repeat for each time step Dt
1. compute direction ci and speed vi r1 r2 rn
2. compute SSF gi
3. compute distance ri to receiver

40. Real-time Sound Rendering
Procedure
- repeat for each time step Dt
1. compute direction ci and speed vi
2. compute SSF gi +
3. compute distance ri to receiver
4. compute sound pressure pv
Curle‘s model

41. Real-time Sound Rendering
Procedure
- repeat for each time step Dt
direction
(c1, c2, …, cn)
speed
(v1, v2, …, vn)
w(l, s, u, v0)
sound texture
values of SSF
(g1, g2, …, gn)
1. compute direction ci and speed vi
2. compute SSF gi
(texture for base speed)
3. compute distance ri to receiver
4. compute sound pressure pv

42. Computation of SSF Property vi freq. ∝ speed v amp. ∝ (speed v )6 Dt t
actual speed
freq. ∝ speed v
amp. ∝ (speed v )6 w
sound texture
(base speed v0) s

43. Computation of SSF Property vi freq. ∝ speed v vi(k)
amp. ∝ (speed v )6 vi
actual speed
vi(k) k Dt t
different interval s w
sound texture
(base speed v0)

44. Computation of SSF Property vi freq. ∝ speed v vi(k)
amp. ∝ (speed v )6 vi
actual speed
vi(k) k Dt t
x(vi(k)/v0)6 s w
vi(k)/v0xDt
sound texture
(base speed v0)

45. Computation of SSF Property Recurrence relation Periodical use vi
freq. ∝ speed v
amp. ∝ (speed v )6 vi
actual speed Dt
Recurrence relation t ï î í ì = + D - ) , ( / 6 1 v s l t k c w g
overlap w s w
Periodical use
blending for smooth transition
sound texture
(base speed v0)

46. Overview Introduction Related Work
Principle and Prediction of Aerodynamic Sound
Basic Idea of Our Method
Computation of Sound Texture
Real-time Sound Rendering
Examples
Conclusions

47. Fluid Simulation Demo
Sound texture for square prism for one direction of flow
length 50cm, side length 2.0cm
base speed 10 m/s
2D analysis
finite difference

48. Real-time Sound Rendering Demo
Rotating sphere
wire has no effect on sound
Doppler effect
Cylinder thrown at receiver
rotating as it approaching
Doppler effect
Sound by wind
wind through fence
draft through gap between windows

49. Application Character animation Bear swinging a huge club
Warrior swinging two different swords
(image by TAITO)

50. Conclusions Sound synthesis of fluid
Real-time rendering of aerodynamic sound
sound texture based on CFD
synthesis of sound waves using Curle‘s model
real-time
New element to improve realistic simulation of virtual environments

51. Acknowledgement As for fluid analysis As for character animation
Atsushi Kunimatu (TOSHIBA Corp., Japan),
Tunemi Takahashi (TOSHIBA Corp., Japan),
Naofumi Shibata (TOSHIBA Info. Systems Corp., Japan)
As for character animation
People of GARAKUTA STUDIO (TAITO Corp., Japan)
People of Project BUJINGAI (TAITO Corp., Japan)