Quality, Privacy, Theory and Design.
Jacob Donley, School of Electrical, Computer and Telecommunications Engineering, University of Wollongong (UOW).
Oculus Invited Talk, June 2017.
Call Girls Service Nagpur Tanvi Call 7001035870 Meet With Nagpur Escorts
Improving Personal Sound Zone Reproductions
1. Improving Personal Sound Zone Reproductions
Quality, Privacy, Theory and Design
Jacob Donley
School of Electrical, Computer and Telecommunications Engineering,
University of Wollongong (UOW)
Oculus Invited Talk, June 2017
2. Overview
1 Introduction
2 Background
3 Providing privacy in video conferences
4 Cancelling speech between people in a shared space
5 Reducing the cost with fewer loudspeakers
6 Conclusions
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 0 / 32
3. Overview
1 Introduction
2 Background
3 Providing privacy in video conferences
4 Cancelling speech between people in a shared space
5 Reducing the cost with fewer loudspeakers
6 Conclusions
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 0 / 32
4. Introduction
Sound and audio reproduction
Localisation
Realism
Sound Fields
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 1 / 32
5. Introduction
Sound and audio reproduction
Localisation
Realism
Sound Fields
Personalised sound
Multilingual Home
Entertainment
Immersive Audio/Video Cinema
Shared Gaming Spaces
3D Audio/Video
Teleconferencing
[1]
[2]
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 1 / 32
6. Overview
1 Introduction
2 Background
3 Providing privacy in video conferences
4 Cancelling speech between people in a shared space
5 Reducing the cost with fewer loudspeakers
6 Conclusions
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 1 / 32
7. A view of soundfield theory
How do we hear?
[3]
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 2 / 32
8. A view of soundfield theory
How do we determine the direction of a sound?
Interaural Time Difference
(ITD) (< 1kHZ)
Phase delay (low frequencies)
Group delay (high
frequencies)
Interaural Level Difference
(ILD) (> 1.5kHZ)
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 3 / 32
9. A view of soundfield theory
How do we determine the direction of a sound?
Interaural Time Difference
(ITD) (< 1kHZ)
Phase delay (low frequencies)
Group delay (high
frequencies)
Interaural Level Difference
(ILD) (> 1.5kHZ)
Pinnae-based Spectral Cues
+30◦
0◦
−30◦
Frequency
LevelDifference
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 3 / 32
10. A view of soundfield theory
How do we determine the direction of a sound?
Interaural Time Difference
(ITD) (< 1kHZ)
Phase delay (low frequencies)
Group delay (high
frequencies)
Interaural Level Difference
(ILD) (> 1.5kHZ)
Pinnae-based Spectral Cues
How good are we at this?
Localisation accuracy:
≈ 1◦ in front
≈ 15◦ to the side
+30◦
0◦
−30◦
Frequency
LevelDifference
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 3 / 32
11. A view of soundfield theory
How to control the sound that enters the ear?
Head-Related Transfer Functions (HRTFs)
Time Differences
Amplitude Panning
Binaural Rendering/Recording
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 4 / 32
12. A view of soundfield theory
How to control the sound that enters the ear?
Head-Related Transfer Functions (HRTFs)
Time Differences
Amplitude Panning
Binaural Rendering/Recording
Sound Field Synthesis (SFS)
Higher-Order Ambisonics (HOA)
Spectral Division Method (SDM)
Wave Field Synthesis (WFS)
Multizone Sound Field Synthesis
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 4 / 32
13. A view of soundfield theory
How to control the sound that enters the ear?
Head-Related Transfer Functions (HRTFs)
Time Differences
Amplitude Panning
Binaural Rendering/Recording
Sound Field Synthesis (SFS)
Higher-Order Ambisonics (HOA)
Spectral Division Method (SDM)
Wave Field Synthesis (WFS)
Multizone Sound Field Synthesis
How to make sure the perceived direction is accurate?
What we hear should match what we see
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 4 / 32
14. A view of soundfield theory
Binaural Rendering using Head Related Transfer Functions (HRTFs)
Highly dependent on an individuals head and ear
Single wave-front, single loudspeaker
cm
[4]
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 5 / 32
15. A view of soundfield theory
HOA, SDM, WFS and other spatial audio techniques
Independent of an individuals head and ear
Many wave-fronts, many loudspeakers
[4]
cm
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 6 / 32
16. A view of soundfield theory
HOA, SDM, WFS and other spatial audio techniques
Independent of an individuals head and ear
Many wave-fronts, many loudspeakers
[4]
m
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 6 / 32
17. Personal sound zone theory
You can target specific soundfields in specific locations by varying the
outputs from an array of loudspeakers surrounding a space [5], [6].
[5]
[6]
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 7 / 32
18. Common loudspeaker setups
0°
ϕL
DL
Rc ϕc
R
D
Du
rzb
rb
β
b
Db
θ
rzq
rq
ϙ
q
Dq
−ϑ
(rl , ϕl )
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 8 / 32
19. Common loudspeaker setups
0°
ϕL
DL
Rc ϕc
R
D
Du
rzb
rb
β
b
Db
θ
rzq
rq
ϙ
q
Dq
−ϑ
(rl , ϕl )
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 8 / 32
20. Personal sound zone theory
Some soundfield synthesis and reproduction techniques
Pressure Matching
Acoustic Contrast Control
Planarity Control
Cylindrical/Spherical Harmonic
Expansion
Orthogonal Basis Expansion
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 9 / 32
21. Personal sound zone theory
Some soundfield synthesis and reproduction techniques
Pressure Matching
Acoustic Contrast Control
Planarity Control
Cylindrical/Spherical Harmonic
Expansion
Orthogonal Basis Expansion
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 9 / 32
22. Personal sound zone theory
Some soundfield synthesis and reproduction techniques
Pressure Matching
Acoustic Contrast Control
Planarity Control
Cylindrical/Spherical Harmonic
Expansion
Orthogonal Basis Expansion
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 9 / 32
23. Personal sound zone theory
Some soundfield synthesis and reproduction techniques
Pressure Matching
Acoustic Contrast Control
Planarity Control
Cylindrical/Spherical Harmonic
Expansion
Orthogonal Basis Expansion
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 9 / 32
24. Personal sound zone theory
Some soundfield synthesis and reproduction techniques
Pressure Matching
Acoustic Contrast Control
Planarity Control
Cylindrical/Spherical Harmonic
Expansion
Orthogonal Basis Expansion
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 9 / 32
25. Personal sound zone theory
Synthesise a desired soundfield as weighted basis functions
S(x; k) =
∑
h
Wh Ph(x; k) (1)
Find loudspeaker weights using harmonic expansion
Wl (k) =
2∆ϕs
iπ
ĎM∑
sm=−ĎM
∑
h
i smexp(i sm(ϕl − ρh))
H
(1)
sm (rl k)
Wh, (2)
W0(k)
W1(k)
·
·
·
·
·
Wl (k)
Wh
Wh
Wh
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 10 / 32
26. Personal sound zone theory
Synthesise a desired soundfield as weighted basis functions
S(x; k) =
∑
h
Wh Ph(x; k) (1)
Find loudspeaker weights using harmonic expansion
Wl (k) =
2∆ϕs
iπ
ĎM∑
sm=−ĎM
∑
h
i smexp(i sm(ϕl − ρh))
H
(1)
sm (rl k)
Wh, (2)
Loudspeaker driving signals
Ql (a, k) = Wl (k) Y (a, k) (3)
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 10 / 32
27. Personal sound zone theory
Synthesise a desired soundfield as weighted basis functions
S(x; k) =
∑
h
Wh Ph(x; k) (1)
Find loudspeaker weights using harmonic expansion
Wl (k) =
2∆ϕs
iπ
ĎM∑
sm=−ĎM
∑
h
i smexp(i sm(ϕl − ρh))
H
(1)
sm (rl k)
Wh, (2)
Loudspeaker driving signals
Ql (a, k) = Wl (k) Y (a, k) (3)
Reproduced sound pressure levels in space
P(sp)
(x; a, k) =
∑
l
Ql (a, k) T(x, ll ; k), (4)
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 10 / 32
28. Overview
1 Introduction
2 Background
3 Providing privacy in video conferences
4 Cancelling speech between people in a shared space
5 Reducing the cost with fewer loudspeakers
6 Conclusions
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 10 / 32
29. Providing privacy in video conferences
Figure: An example of the multizone soundfield occlusion problem.
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 11 / 32
30. Privacy and Quality Control
How to increase privacy between two spaces?
Define a Joint Speech & Masker Soundfield
P(sp,m)
(x; a, k) = P(sp)
(x; a, k) + GP(m)
(x; a, k) (5)
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 12 / 32
31. Privacy and Quality Control
How to maximise the privacy and quality between two areas?
Speech Intelligibility Contrast (SIC) [7]
SICM = db
−1
ż
Db
IM(p(x; ·); y) dx − dq
−1
ż
Dq
IM(p(x; ·); y) dx (6)
Intelligibility Intelligibility
Privacy (SIC)
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 13 / 32
32. Privacy and Quality Control
How to maximise the privacy and quality between two areas?
Speech Intelligibility Contrast (SIC) [7]
SICM = db
−1
ż
Db
IM(p(x; ·); y) dx − dq
−1
ż
Dq
IM(p(x; ·); y) dx (6)
Privacy and Quality Maximisation [7]
arg max
G
(
SICM +
λ
db
ż
Db
B ´M dx
)
(7)
Intelligibility Intelligibility
Privacy (SIC)
Quality
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 13 / 32
33. Spatial and Spectral Sound Masking
What sound masking magnitude spectrum to use?
Spectra to consider:
Speech
Speech in quiet zone
Masker in bright zone
Spatial aliasing
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 14 / 32
34. Spatial and Spectral Sound Masking
What sound masking magnitude spectrum to use?
Spectra to consider:
Speech
Speech in quiet zone
Masker in bright zone
Spatial aliasing
Intelligibility and Quality Control Filter
H(IB)
(k) = H(sp)
(k)
H(q′)(k)
1−λ`
H(b′)(k)
λ`
, (8)
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 14 / 32
35. Speech Privacy and Quality in Reproductions
Figure: Real-world multizone implementations.
Semi-circular array on top.
Linear array on bottom.
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 15 / 32
36. Speech Privacy and Quality in Reproductions
Bright Zone Intelligibility
Quiet Zone Intelligibility
Bright Zone Quality
Semi-Circle Array Line Array
Simulation
Real-World
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 16 / 32
37. Overview
1 Introduction
2 Background
3 Providing privacy in video conferences
4 Cancelling speech between people in a shared space
5 Reducing the cost with fewer loudspeakers
6 Conclusions
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 16 / 32
38. Cancelling speech between people in a shared space
0°
sR
sϕ
RD
D
rc
Dc
(rt, θt) sD
(rl , ϕl )
Figure: Active
control layout for a
linear dipole array
and speech prediction
microphone. [8]
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 17 / 32
39. Cancelling speech between people in a shared space
0°
sR
sϕ
RD
D
rc
Dc
(rt, θt) sD
(rl , ϕl )
Figure: Active
control layout for a
linear dipole array
and speech prediction
microphone. [8]
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 17 / 32
40. Soundfield Control Technique
Define a control soundfield (sum of weighted basis functions)
Sc
(x; k) =
∑
g
Eg,mFg(x; k) (9)
Find weights that minimise the residual energy
min
Eg,m
∥
∑
g
Eg,mFg(x; k) + St
(x; k)∥2
(10)
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 18 / 32
41. Loudspeaker Weights
Use cylindrical harmonic expansion (2) to determine the monopole
loudspeaker weights, Wl (k).
Model dipoles to reproduce on one side of the array.
Cardioid
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 19 / 32
42. Loudspeaker Weights
Use cylindrical harmonic expansion (2) to determine the monopole
loudspeaker weights, Wl (k).
Model dipoles to reproduce on one side of the array.
Huygens-Fresnel principle
Not strictly Kirchoff-Helmholtz integral
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 19 / 32
43. Loudspeaker Weights
Use cylindrical harmonic expansion (2) to determine the monopole
loudspeaker weights, Wl (k).
Model dipoles to reproduce on one side of the array.
Huygens-Fresnel principle
Not strictly Kirchoff-Helmholtz integral
Dipole loudspeaker weights [8]
Wl,s(k) Wl (k)
exp
(
i(−1)s(k¨d − π)/2
)
2k¨d
(11)
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 19 / 32
44. Autoregression (AR) Parameter Estimation
Soundfield filtering induces inherent delay.
Can we predict the signal ahead of time?
Estimate paj using known past samples [8]
ϵ(n + `b + 1) = v(n + `b + 1) +
∑
j
paj v(n + `b − j) (12)
Autocorrelation method1 gives stable AR coefficients, paj .
1
Equivalent to the Yule-Walker method
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 20 / 32
45. AR Filter Delay Compensation
Use estimated parameters to forecast signal
v(n + ´b + 1) = −
∑
j
paj v(n + ´b − j), ∀´b ∈ pM (13)
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 21 / 32
46. AR Filter Delay Compensation
Use estimated parameters to forecast signal
v(n + ´b + 1) = −
∑
j
paj v(n + ´b − j), ∀´b ∈ pM (13)
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 21 / 32
47. Soundfield Suppression
Figure: 1kHz pressure field.
64ms latency.
Inactive (A).
Active (B).
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 22 / 32
48. Synthesis and Prediction Accuracy Trade-Off
4 8 12 16 20 24 28 32
Block Length (ms)
-20
-15
-10
-5
0
Suppression(dB)
Predicted Signal Actual Signal
Figure: Mean suppression for an
actual future block and
predicted future block.
0.1 1 8
Frequency (kHz)
-15
-10
-5
0
5
Suppression(dB)
Predicted Signal Actual Signal
Figure: Suppression for a 12 ms
block length.
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 23 / 32
49. Overview
1 Introduction
2 Background
3 Providing privacy in video conferences
4 Cancelling speech between people in a shared space
5 Reducing the cost with fewer loudspeakers
6 Conclusions
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 23 / 32
50. Reducing the cost with fewer loudspeakers
1.7kHz 2.5kHz 5.0kHz
Figure: An example of multizone spatial aliasing.
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 24 / 32
51. Modelling Spatial Aliasing
0°
DL
Rc ϕc
R
R′
rzb
rb
β
b
θ
rzq
rq
ϙ
q
α
Ĺpb
p
Ĺpq
“γ−
rb
d⊥
“gu
d⊥
Ĺpb
“g−
u
Figure: Auxiliary entities:
Circular array.
Plane-wave vector in blue.
Grating lobe limit in red.
Frequency limit computed with
values in green.
ku =
2π(L − 1) − ϕL
(
d⊥
“gu
+ d⊥
Ĺpb
)
ϕL
(14)
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 25 / 32
52. Modelling Spatial Aliasing
0°
DL
Rc ϕc
R
R′
rzb
rb
β
b
θ
rzq
rq
ϙ
q
Ĺpb
p
Ĺpq
sγ
rb
sg−
u
Figure: Auxiliary entities:
Linear array.
Plane-wave vector in blue.
Grating lobe limit in red.
Frequency limit computed with
values in green.
ku =
2π(L − 1)
DL(sin(sγ − Θ) + sin(Θ))
(15)
where Θ is a rotation invari-
ant array angle
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 25 / 32
53. Reducing the cost with fewer loudspeakers
Add weight to multizone soundfield
Sa
MSR(x, k) = GMSR(k)
∑
l
Wl (k)T(x, ll , k) (16)
Add weight to parametric loudspeaker soundfield
Sa
PL(x, k) = GPL(k)E(x, k)D(x, k)eik∥x−p∥
(17)
Figure: Parametric loudspeakers [9], [10].
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 26 / 32
54. Cross-over Filter
Low-pass and high-pass
Linkwitz-Riley filters
GMSR(k) = Bˆn
2
(k/ku)−2
(18)
GPL(k) = Bˆn
2
(ku/k)−2
(19)
Flat frequency response
|GMSR(k) + GPL(k)| = 1 (20)
Frequency
Gain
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 27 / 32
55. Cross-over Filter
Low-pass and high-pass
Linkwitz-Riley filters
GMSR(k) = Bˆn
2
(k/ku)−2
(18)
GPL(k) = Bˆn
2
(ku/k)−2
(19)
Flat frequency response
|GMSR(k) + GPL(k)| = 1 (20)
Frequency
Gain
Hybrid synthesised soundfield
Sa
H(x, k) =
∑
R
db|GR(k)|Sa
R(x, k)
ş
Db
Sa
R(x, k) dx
(21)
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 27 / 32
56. Acoustic Contrast Improvement
0
20
40
60
80
100
120
140
AcousticContrast(dB)
- 50
- 40
- 30
- 20
- 10
0
MeanSquaredError(dB)
L = 24
0. 1 1 8
L = 24
L = 134
0. 1 1 8
Frequency (kHz)
L = 134
Figure: Acoustic Contrasts and Spatial Errors.
L = 134 is alias free up to 8 kHz.
Multizone Soundfield
Reproduction (MSR)
Parametric
Loudspeaker (PL)
Hybrid (H)
Aliasing frequency (ku)
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 28 / 32
57. Overview
1 Introduction
2 Background
3 Providing privacy in video conferences
4 Cancelling speech between people in a shared space
5 Reducing the cost with fewer loudspeakers
6 Conclusions
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 28 / 32
58. Conclusions
Improved video conferencing using perceptually weighted masking
Improved shared spaces from cross-zone speech cancellation
(e.g. gaming, conferencing, cinema)
Cost effective installations
Reduced loudspeaker counts
Zone-based spatial aliasing
Parametric loudspeakers
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 29 / 32
59. Future Work
Some gaps in the current knowledge:
Unified theory (privacy, quality, cancellation and loudspeaker
reduction)
Joint optimisation of cost functions
De-reverberation with no intrusive microphones
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 30 / 32
60. References: I
[1] Gramophone Maryland, Home Theater, Mar. 2010. [Online]. Available:
https://www.flickr.com/photos/gramophonemaryland/5506863384/.
[2] Fuelrefuel, Teliris VirtuaLive Telepresence Modular System, 2007. [Online]. Available:
https://commons.wikimedia.org/wiki/File:Teliris_VL_Modular.JPG.
[3] C. L. Brockmann, A diagram of the anatomy of the human ear, Feb. 2009. [Online].
Available:
https://commons.wikimedia.org/wiki/File:Anatomy_of_the_Human_Ear_en.svg.
[4] Vector graphics created by Freepik, Jun. 2017. [Online]. Available: www.freepik.com.
[5] J. Donley and C. Ritz, Just for you: how to create sounds that only you can hear in a
venue. The Conversation, 2016.
[6] T. Betlehem, W. Zhang, M. Poletti, and T. D. Abhayapala, “Personal Sound Zones:
Delivering interface-free audio to multiple listeners,” IEEE Signal Process. Mag.,
vol. 32, pp. 81–91, 2015.
[7] J. Donley, C. Ritz, and W. B. Kleijn, “Improving speech privacy in personal sound
zones,” in Int. Conf. on Acoust., Speech and Signal Process. (ICASSP), IEEE, 2016,
pp. 311–315.
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 30 / 32
61. References: II
[8] J. Donley, C. Ritz, and W. B. Kleijn, “Active speech control using wave-domain
processing with a linear wall of dipole secondary sources,” in Int. Conf. on Acoust.,
Speech and Signal Process. (ICASSP), IEEE, 2017, pp. 1–5.
[9] C. Shi and W.-S. Gan, “Development of Parametric Loudspeaker,” IEEE Potentials,
vol. 29, no. 6, pp. 20–24, Nov. 2010.
[10] Yongsheng Mu, Peifeng Ji, Wei Ji, Ming Wu, and Jun Yang, “Modeling and
Compensation for the Distortion of Parametric Loudspeakers Using a One-Dimension
Volterra Filter,” IEEE/ACM Transactions on Audio, Speech, and Language
Processing, vol. 22, no. 12, pp. 2169–2181, Dec. 2014.
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 31 / 32
65. Appendix: Geometric Delay Compensation
Microphone signal is attenuated and time-delayed.
Inverse filter to "virtual sense" talker signal
Geometric Delay Compensation [8]
v(n) = Re
{
1
N
∑
m
4
{∑
n z(n)exp
(
−icnkm/2˙f
)}
iH
(1)
0 (km ∥v − z∥)
exp
(
icnkm/2˙f
)
}
(25)
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 32 / 32
66. Appendix: Directivity Models
Parametric loudspeaker soundfield computed from:
Directivity coefficients
E(x, k) =
˜βk2
4π ˜αs ˜ρ0 x − p c2
(26)
Convolutional directivity
D(x, k) = [DG(x, kc)DG(x, kc + k)] ∗ DW(x, k) (27)
Gaussian directivity
DG
(
x, ˆk
)
= e( i
2
dˆk tan (ρx+Ψ))
2
(28)
Westervelt’s directivity
DW(x, k) = ˜αs/
b
˜α2
s + k2 tan4 (ρx + Ψ) (29)
Jacob Donley (UOW) Personal Sound Zones Oculus 2017 32 / 32