#2015 The Acoustical Society of Japan
Acoust. Sci. & Tech. 36, 6 (2015)
Development of dynamic crosstalk cancellation system
for multiple-listener binaural reproduction
Hiroaki Kurabayashi1 , Makoto Otani2; , Masami Hashimoto3 and Mizue Kayama3
1
Graduate School of Science and Technology, Shinshu University,
4–17–1 Wakasato, Nagano, 380–8553 Japan
2
Graduate School of Engineering, Kyoto University,
Kyoto daigaku-Katsura, Nishikyo-ku, Kyoto, 615–8540 Japan
3
Faculty of Engineering, Shinshu University,
4–17–1 Wakasato, Nagano, 380–8553 Japan
(Received 12 February 2015, Accepted for publication 30 March 2015)
Keywords: Binaural reproduction, Multiple listeners, Crosstalk cancellation, Head tracking
PACS number: 43.66.Pn [doi:10.1250/ast.36.537]
1.
Introduction
TransauralÒ y reproduction [1,2] is one approach to realize
binaural presentation of auditory scenes to a listener using
loudspeakers and crosstalk cancellation (CTC), instead of a
set of headphones. Wireless and non-contact binaural reproduction achieved by transaural reproduction is the approach’s
main advantage over headphone presentation.
One shortcoming of conventional transaural reproduction
is that CTC filters vary depending on the listener’s head
orientation and position. Therefore, if static CTC filters are
used, then the listener is forced to remain motionless during
listening. However, recent research has indicated that dynamic transaural reproduction using a head tracking device
and adaptive CTC filter processing provides accurate binaural
presentation even when a listener rotates his/her head [3–5].
Another concern is that most conventional transaural reproduction systems were developed to present binaural signals to
the ears of a single listener, thereby disabling its use by
multiple listeners. In theory, simultaneous binaural reproduction to multiple listeners by transaural reproduction is
realizable using crosstalk cancellation assuming multiple
listeners, but development of such multi-listener transaural
reproduction is insufficient [6,7]. No attempt has been
reported to apply multi-listener crosstalk cancellation to
dynamic transaural reproduction.
This paper presents the development of a dynamic
transaural reproduction system for multiple listeners, based
on a single-listener dynamic transaural reproduction system
that was developed previously [5]. This report describes
results of a subjective evaluation of localization performance
using the developed system.
2.
Multi-listener dynamic transaural reproduction
system
A dynamic transaural reproduction system for two
listeners was developed as an extension of a single-listener
e-mail: otani@archi.kyoto-u.ac.jp
Transaural is a trademark of Cooper Bauck Corporation.
y
system developed previously by the authors [5]. Figure 1
presents a schematic of the multi-listener dynamic transaural
reproduction system. Pure Data (Pd) on Apple OSX (Apple
Computer Inc.) was used to operate real-time generation of
CTC filters and convolution. Based on the head position and
orientation of each listener detected by Microsoft Kinect
operating on Windows OS (Microsoft Corp.), Pd synthesizes
binaural signals that are to be reproduced at each listener’s
ears, generates CTC filters, and convolves source signals to
head-related transfer functions (HRTFs) and CTC filters.
OpenSound Control (OSC) protocol [8] on UDP (User
Datagram Protocl) is used for data communication between
Windows and OSX PCs. The CTC filters are obtained in
real time by least-norm-solution in a frequency domain [9,10]
to reproduce the synthesized binaural signals at four control
points: four ears of two listeners. In CTC filter design, the
filter length is 1,024 pt. The regularization parameter is
0.000,1 for all frequencies. Acoustic influences between
two listeners are neglected. An HRTF database used for both
binaural synthesis and CTC filter generation was obtained
from the boundary element simulation [11] and a computer
head and pinnae model that was captured by magnetic
resonance imaging. The HRTFs were prepared for point
sources located between 0.15 m and 2 m from the center of
the head with 0.05-cm intervals and 5-deg intervals of
azimuth and elevation. The system includes a loudspeaker
array consisting of 24 loudspeaker units with vibration
surfaces facing upward to simulate omni-directivity in the
horizontal plane.
A measurement showed that signal processing delay is
approximately 9 ms for both CTC filter generation and
convolution of CTC filter and binaural signals, respectively.
Assuming approximately 100 ms delay of Kinect and other
delays, the total system latency would exceed 100 ms, which
is larger than the acceptable system latency of a virtual
auditory display [12]. Such large latency might degrade the
presentation of binaural signals and perceived auditory space
when listeners move quickly. This issue shall be addressed in
future works.
537
Pure Data
Listeners
Power Amplifier
(MBA-32, J.TESORI)
24ch array speakers
Audio I/F
(828mk3 Hybrid, MOTU)
DAC
(ADA8000, BEHRINGER)
60
*
**
50
40
30
20
10
0
3.9 [cm] intervals,
24 loudspeakers
1 [m]
1 [m]
0.4 [m]
0.8 [m]
0.4 [m]
70
60
50
40
30
20
10
0
Listener
(a) Center.
Imaginary
Listener
Listener
Imaginary
Listener
(b) Left.
Fig. 2 Experimental conditions: (a) center and (b) left
listening positions.
3. Sound localization experiment
3.1. Method
Six subjects participated in a sound localization experiment performed in an anechoic chamber at the Faculty of
Engineering, Shinshu University. Figure 2 presents configurations of a line loudspeaker array, Kinect, and listeners. The
line loudspeaker array consisted of 24 loudspeaker units (
25 mm, NSW1-205-8A; Aurasound) installed to a wooden
frame (1;143 mm 87 mm 38 mm) at 39-mm intervals. To
eliminate visual cues, the loudspeaker array was veiled by a
black cloth that was confirmed to be acoustically transparent
before the experiment. Other equipment is presented in Fig. 1.
One non-individualized set of HRTFs was used for all
subjects. The stimulus was pink noise of 3-s duration. Aweighted sound pressure level was approximately 61 dB when
a sound image was presented from the frontal direction and
1-m distance.
Sound images were presented at 12 horizontal directions
with 30-deg. intervals. Trials were performed by single
subjects, but the transaural system was operated assuming
two listeners so that same sound images were presented to
the second ‘‘imaginary’’ listener. Each stimulus for one sound
image position was presented five times. Therefore one
session consisted of 60 (¼ 5 12) trials in all, which were
presented in a randomized order. The session was performed
for three conditions: Static, Non-tracking, and Dynamic. In
the Static condition, subjects were instructed not to move their
head during listening to each stimulus. In Non-tracking and
Dynamic conditions, subjects were instructed to rotate their
head from the front to left (30 deg), right (30 deg), and then
538
40
30
20
10
13.1 15.9 16.2
Upper column: back to front
Lower column: front to back
80
60
**
**
40
20
0
: Non-tracking
LR Average angular errors [deg]
3.9 [cm] intervals,
24 loudspeakers
50
100
36.0 45.0 8.3
: Dynamic
(a) Center.
Average angular errors [deg]
0.1 [m]
Kinect
60
0
41.6 53.7 19.2
80
0.1 [m]
70
: Static
Fig. 1 Schematic of the multi-listener dynamic transaural reproduction system.
Kinect
80
51.9 63.9 45.0
: Static
80
70
60
50
40
30
20
10
0
19.3 23.9 27.6
: Non-tracking
Average front-back reversal ratio [%]
OSX PC
(MacBook Pro, Apple)
Average angular errors [deg]
Kinect App
OSC
on
UDP
70
Average front-back reversal ratio [%]
80
Kinect
Windows PC
(Probook 4340s, HP)
LR Average angular errors [deg]
Acoust. Sci. & Tech. 36, 6 (2015)
100
Upper column: back to front
Lower column: front to back
80
60
*
40
20
0
42.1 47.7 24.7
: Dynamic
(b) Left.
Fig. 3 Experimental results: Averaged angular errors,
left-right (LR) angular errors, and front-back reversal.
to the front while listening to each stimulus. In the Dynamic
condition, the system responded to the subject’s head rotation,
although it did not in the Non-tracking condition. In all the
conditions, the imaginary listener was assumed to be still.
The experiment was performed for two listening positions,
as presented in Fig. 2: participants were located on the center
or left side of the loudspeaker array to assess the effect of
listening position relative to the loudspeaker array.
3.2. Results and discussion
Figures 3(a) and 3(b), respectively portray experimental
results for the center and left listening positions. The left,
center, and right panels in each figure respectively represent
an angular error [deg], left–right (LR) angular error [deg], and
front–back reversal ratio [%], which are averaged among all
subjects. The angular error denotes an absolute error between
the presented and answered angles. The LR angular error
represents an absolute error by which the answered angle is
reversed with respect to the subject’s transversal plane when
front–back revearsal occurs. For the front–back reversal ratio,
upper and lower columns respectively show front-to-back and
back-to-front misjudgements. Asterisks () denote the result
of multiple comparison (Tukey method) among all conditions
(: p < 0:05, : p < 0:01).
For the center listening position depicted in Fig. 3(a), the
angular error and front–back reversal ratio are significantly
smaller in the Dynamic condition than in other conditions,
although no significant difference was found in LR angular
error among the conditions. Especially, the front–back
revearsal ratio is greatly reduced to less than 10% in the
Dynamic condition, which indicates that the dynamic trans-
H. KURABAYASHI et al.: MULTI-LISTENER DYNAMIC CROSSTALK CANCELLATION
aural reproduction functions successfully even when applied
to a multi-listener use.
However, for the left listening position depicted in
Fig. 3(b), localization errors are greater than those in the
center listening position, although the front–back reversal
ratio is significantly smaller in the Dynamic condition than in
Non-tracking condition, which indicates that, for the experimental setup of the current study, the performance of the
multi-listener dynamic transaural reproduction depends on
the listening position. Such degradation would be attributable
to an increased error of Kinect tracking that might be
prominent when a listener is located close to an edge of
Kinect’s field of view or to an increased error of binaural
signals reproduced at both a listener’s ears caused by
inadequate positional relationship between loudspeakers and
both a listener’s ears leading to ill-conditioned CTC filter
generation.
4.
Summary
This paper presents a development of dynamic transaural
reproduction system for multiple listener. A sound localization experiment was conducted to evaluate the performance of the developed system subjectively. The results reveal
that the multi-listener dynamic transaural reproduction functions properly when a listener is located immediately in front
of the loudspeaker array, whereas showing a degraded
performance when a listener is located off-center. Such
degradation would be eliminated by improvements in the
tracking sensor or loudspeaker arrangement, which should be
addressed in future works.
Acknowledgment
This work was partially supported by a Grant-in-Aid for
Scientific Research (B) (No. 26280078), MEXT, Japan.
References
[1] M. R. Schroeder and B. S. Atal, ‘‘Computer simulation of
sound transmission in rooms,’’ IEEE Conv. Rec., 7, 150–155
(1963).
[2] D. H. Cooper and J. L. Bauck, ‘‘Prospects for transaural
recording,’’ J. Audio Eng. Soc., 37, 3–19 (1989).
[3] W. G. Gardner, 3-D Audio Using Loudspeaker, Massachusetts
Institute of Technology, Ph.D. Thesis (1997).
[4] T. Lentz, ‘‘Dynamic crosstalk cancellation for binaural synthesis in virtual reality environment,’’ J. Audio. Eng. Soc., 54,
283–294 (2006).
[5] H. Kurabayashi, M. Otani, K. Itoh, M. Hashimoto and M.
Kayama, ‘‘Sound image localization using dynamic transaural
reproduction with non-contact head tracking,’’ IEICE Trans.,
E97-A, 1849–1858 (2014).
[6] Y. Kahana, P. A. Nelson, O. Kirkeby and H. Hamada,
‘‘Objective and subjective assessment of systems for the
production of virtual acoustic images for multiple listeners,’’
Audio Eng. Soc. 103rd Conv., pre-print 4573 (1997).
[7] Y. Kim, O. Deille and P. A. Nelson, ‘‘Crosstalk cancellation
in virtual acoustic imaging systems for multiple listeners,’’
J. Sound Vib., 297, 251–266 (2006).
[8] OpenSound Control: http://opensoundcontrol.org/ (accessed
2015-09-25).
[9] A. Kaminuma, Wide Area Sound Field Reproduction System
Design Using Inverse Filter, Nara Institute of Science and
Technology, Ph.D. Thesis (2001).
[10] A. Tanaka, N. Otani and M. Miyakoshi, ‘‘Regularized inverse
filter design for sound reproduction system,’’ IEICE Trans.,
J87-A, 1466–1467 (2004).
[11] M. Otani and S. Ise, ‘‘Fast claculation system specialized for
head-related transfer function based on boundary element
method,’’ J. Acoust. Soc. Am., 119, 2589–2598 (2006).
[12] S. Yairi, Y. Iwaya and Y. Suzuki, ‘‘Estimation of detection
threshold of system latency of virtual auditory display,’’ Appl.
Acoust., 68, 851–863 (2007).
539