Simulated head related transfer function of the phyllostomid bat Phyllostomus discolor

F. De Mey, J. Reijniers, H. Peremans, M. Otani, and U. Firzlaff

Citation: The Journal of the Acoustical Society of America 124, 2123 (2008); doi: 10.1121/1.2968703
View online: https://doi.org/10.1121/1.2968703
View Table of Contents: https://asa.scitation.org/toc/jas/124/4
Published by the Acoustical Society of America

ARTICLES YOU MAY BE INTERESTED IN

The bat head-related transfer function reveals binaural cues for sound localization in azimuth and elevation
The Journal of the Acoustical Society of America 116, 3594 (2004); https://doi.org/10.1121/1.1811412

A numerical study of the role of the tragus in the big brown bat
The Journal of the Acoustical Society of America 116, 3701 (2004); https://doi.org/10.1121/1.1815133

Numerical calculation of listener-specific head-related transfer functions and sound localization: Microphone
model and mesh discretization
The Journal of the Acoustical Society of America 138, 208 (2015); https://doi.org/10.1121/1.4922518

Boundary element method calculation of individual head-related transfer function. I. Rigid model calculation
The Journal of the Acoustical Society of America 110, 2440 (2001); https://doi.org/10.1121/1.1412440

The role of pinna movement for the localization of vertical and horizontal wire obstacles in the greater horseshoe
bat, Rhinolopus ferrumequinum
The Journal of the Acoustical Society of America 84, 1676 (1988); https://doi.org/10.1121/1.397183

Spatially dependent acoustic cues generated by the external ear of the big brown bat, Eptesicus fuscus
The Journal of the Acoustical Society of America 98, 1423 (1995); https://doi.org/10.1121/1.413410
Simulated head related transfer function of the phyllostomid
bat Phyllostomus discolor
F. De Mey,a兲 J. Reijniers, and H. Peremans
Department Milieu, Technologie en Technologiemanagement, Universiteit Antwerpen, Prinsstraat 13, 2000
Antwerpen, Belgium

M. Otani
Research Institute of Electrical Communication, Tohoku University, 2-1-1 Katahira, Aoba-ku,
Sendai 980-8577, Japan

U. Firzlaff
Department Biologie II, Ludwig-Maximilians-Universität München, Grosshaderner Strasse 2,
82152 Planegg-Martinsried, Germany

共Received 14 June 2007; revised 6 June 2008; accepted 16 July 2008兲

This paper presents a calculation of the head related transfer function 共HRTF兲 for the frontal
hemisphere of the phyllostomid bat Phyllostomus discolor using an acoustic field simulation tool
based on the boundary element method. From the calculated HRTF results, binaural interaural
intensity differences 共IIDs兲 are derived. The results: Region of highest sensitivity, HRTF patterns,
and IID patterns are shown to be in good agreement with earlier experimental measurements on
other specimens of the same bat species, i.e., the differences are within the interspecies variability
range. Next, it is argued that the proposed simulation method offers distinct advantages over
acoustic measurements on real bat specimens. To illustrate this, it is shown how computer
manipulation of the virtual morphology model allows a more detailed comprehension of bat spatial
hearing by investigating the effects of different head parts on the HRTF. From this analysis it is
concluded that for this species the pinna has a significantly larger effect on the HRTF and IID
patterns than the head itself. This conclusion argues in favor of a series of recent simulation studies
based on pinna morphology only 关R. Müller, J. Acoust. Soc. Am. 116, 3701–3712 共2004兲; Müller
et al., ibid 119, 4083–4092 共2006兲兴.
© 2008 Acoustical Society of America. 关DOI: 10.1121/1.2968703兴
PACS number共s兲: 43.64.Ha, 43.80.Ka, 43.20.Fn 关MCH兴 Pages: 2123–2132

I. INTRODUCTION importance of binaural IID cues for humans14 as well as for

Bats rely on echolocation for navigation and foraging. In
the search for auditory localization cues available to the bat
during echolocation, the monaural head related transfer func-
tion 共HRTF兲 and the binaural interaural intensity differences
共IIDs兲 have been extensively studied. The monaural HRTF
characterizes the frequency-dependent relation between the
sound pressure of sources at various positions in the free
field and the pressure received at the eardrum.1 This relation
is determined by absorption, diffraction, and reflection of the
sound waves on the head and the external ears 共pinnae兲 of
the subject. Experiments on bats,2–6 humans,7,8 and other
mammals9–11 indicate that monaural HRTF information is
necessary for elevation estimation of the sound source. How-
ever, Aytekin et al.12 conclude that while monaural informa-
tion contains mostly cues for elevation discrimination, espe-
cially in the vertical midplane, it contains ambiguous
azimuthal information as well. Nevertheless, to acquire azi-
muthal information, bats are known to rely predominantly on
binaural IID cues.13 In addition, several studies show the
a兲
Electronic mail: fons.demey@ua.ac.be
other animals,1,9,15 in particular for sound localization at high
frequencies.
Because of their importance for spatial hearing, both the
HRTFs and IID patterns have been measured for different bat
species2,4,5,12,16 as well as for other animals.9–11,17 HRTF
measurements are done by inserting a microphone in the ear
canal at the eardrum’s position and recording sound pulses
sent from several locations in the free field around the head.
These measurements are laborious and great care needs to be
taken in the setup: An anechoic chamber is needed, micro-
phone calibration is necessary, and special specimen prepa-
ration should be performed. To keep the duration of the mea-
surements within reasonable bounds, the number of azimuth
and elevation positions considered is usually limited.1,8,17
Recently, acoustic simulations based on detailed mor-
phological models have been proposed to address the meth-
odological problems inherent to the measurement process.
Such simulations have been performed to obtain the HRTF
for human heads using boundary element methods
共BEM兲18–22 and finite difference methods,23 and for bat pin-
nae using finite element methods 共FEM兲.24,25 Since BEM re-
quires only the boundaries of the model to be discretized,
while for FEM discretization of the whole computational do-

J. Acoust. Soc. Am. 124 共4兲, October 2008 0001-4966/2008/124共4兲/2123/10/$23.00 © 2008 Acoustical Society of America 2123
main is necessary,26 the experiments in this paper are ob-
tained with BEM. While this is numerically more challeng-
ing, computation will consume less memory.
While valid physical arguments can be made for using
models of the pinnae only,25 analyzing the sound field around
a complete head, as proposed in this paper, has some distinct
advantages: True binaural IID data can be obtained and the
influence of the head and various head parts on the HRTF/
IID can be studied. Hence, the main contributions of this
paper are as follows:
共1兲 Comparing HRTF and IID data calculated with the simu-
lation software to HRTF and IID data obtained in mea-
surements allows the examination of the viability of the
simulation approach as well as the validity of the as-
sumptions made in this approach, for a complete head. FIG. 1. Shadow image resulting from a microCT scan. Hairs are not visible.
共2兲 The influence of different head parts is assessed, both on Inset: 3D model, positioned corresponding to shadow image.
the monaural HRTF and the binaural IID patterns.
ning a complete head, these problems complicate both the
An interesting result from this last study, providing fur- reconstruction and model creation steps. First, the calcium
ther evidence in favor of pinna only studies, is that from all present in the skull of the bat has a high x-ray absorption
the head parts, the pinnae have the single biggest effect on ratio compared to the pinna and head tissue. Second, the
HRTF/IID. hairs present on the head are not easily captured and pro-
After an explanation of the methods and techniques used cessed 共see Fig. 1兲. Third, drying artifacts are more difficult
in Sec. II, the comparison between the measured and the to avoid due to increased scanning times 共⬃2 h for the head
simulated bat HRTF and IID patterns is presented in Sec. III. used in this paper兲. Scanning a single pinna on the same
The influence of the different head parts on the bat’s spatial machine would require 45 min. Finally, data sets tend to
hearing is discussed in Sec. IV, followed by a conclusion in grow large 共2 Gbytes for the head used in this paper兲.
Sec. V.
B. Model simplification
II. METHODS AND TECHNIQUES
The head model was simplified with standard mesh sim-
The following outlines the methods and techniques used plification algorithms to reduce the initial model’s complex-
throughout the calculation process: from model acquisition ity 共⬇106 triangles兲 to a manageable level 共⬇30 000 tri-
to result visualization. angles兲. A maximal error of 0.5 mm was introduced, which
was the size at which the simplification process produced a
A. Model acquisition model with around 30 000 surface elements. Moreover, the
maximal edge length was varied adaptively over the head: It
A model of the Phyllostomus discolor head was acquired
was 0.85 mm on the head and on the contralateral pinna 共the
by scanning a real head on a microCT machine. One of the
pinna in which we did not position a receiver兲, 0.68 mm in
pinnae could be scanned in its natural position. Unfortu-
the vicinity of the ipsilateral pinna 共the pinna in which we
nately the other one got slightly bent due to transportation.
placed the receiver兲 and on the back side of that pinna, and
However, the deformation was not so severe that it would
0.56 mm on the sound exposed side of the ipsilateral pinna.
significantly alter the transfer function of the head for a re-
This way it was guaranteed that at the highest calculated
ceiver in the good ear.
frequency the simulation would sample the wave on the
The head was fixated and preserved in formaldehyde
boundary with at least four points per wavelength away from
and dried for approximately 20 min before scanning it with a
the receiver, going to six points per wavelength when ap-
Skyscan27 1076 microCT machine, at a resolution of
proaching the receiver. Numerical consistency checks show
35.36 ␮m. The resulting shadow images were processed
no significant discretization effects on the results when stay-
with cone-beam reconstruction software that accompanies
ing within these bounds.
the scanner, using a Feldkamp reconstruction algorithm. This
resulted in a tomographic representation of the head, which
C. Simulation
was then used to create a stl-boundary 共StereoLithography兲
representation, containing over 1 million boundary elements. For the purposes of this paper it is required that a com-
Limitations to the simulation environment required an addi- puter model of a complete bat head is analyzed. The size of
tional model simplification step 共see Sec. II B兲. a complete head 共35⫻ 35⫻ 35 mm3兲 is large compared to the
MicroCT is well suited for capturing high density mate- wavelength at the maximum simulated frequency 共f
rials, but suffers from some problems, i.e., contrast problems = 95 kHz, ␭ = 3.6 mm兲. Hence BEM was chosen as the most
and drying artifacts, when imaging soft, wet tissue. They can appropriate numerical technique. Nevertheless, the head
be avoided or at least minimized by limiting oneself to only models created, as explained earlier, using standard rules of
scanning the ears, as done in Refs. 24 and 25. When scan- thumb for BEM maximum element size, i.e., having bound-

2124 J. Acoust. Soc. Am., Vol. 124, No. 4, October 2008 De Mey et al.: Calculating the head related transfer function

FIG. 2. MicroCT slice on which the left ear of the bat is shown. The outline
of the boundary model is superimposed on it 共black line兲. Five receivers at
different locations along the ear canal are indicated with a dot. The canal
approximately extends to the tympanum. Inset: Five points on a plane per-
pendicular to the canal composing a receiver.
ary elements with a maximum edge length of 1 / 4th to 1 / 6th
of the wavelength at the maximum frequency,18,20 still chal-
lenge present computing capabilities.
The simulation of the HRTF was performed in the fre-
quency domain, using the BEM3D software developed to con-
duct HRTF calculations on models of the human head, de-
scribed in Refs. 20 and 21. Given the expected complexity of
head models, it was designed to be computationally more
efficient than the traditional BEM methods.
The software allows one to introduce different acoustic
impedance properties per boundary element. However, since,
to the best of our knowledge, no impedance data exist for fur
covered skin in the relevant frequency band, a rigid bound-
ary model was used. This will affect the HRTF simulation
results19 and it is suggested in Sec. III that this approxima-
tion might be responsible for the reduced dynamic range of
the calculated HRTFs when compared to the measured ones.
The simulation itself was performed on the CalcUA-
computer, a cluster containing 256 AMD dual opteron nodes.
Only the nodes with 8 Gbytes of RAM memory at their dis-
posal, however, could be used. On those nodes, it was pos-
sible to simulate models with up to ⬃32 000 triangles.
Since the simulation results are compared to measured
HRTFs of the Phyllostomus discolor,2 the HRTF is simulated
between 25 and 95 kHz, the same range used in these mea-
surements. In Ref. 28 it is shown that this range contains the
frequencies for which the Phyllostomus discolor is most sen-
sitive. Moreover, most of the echolocation call’s energy is
located between 45 and 100 kHz,2,28 indicating that the main
echolocation frequency range is captured by the simulations.
The point receiver used in the simulations is actually a
virtual point receiver. The sound pressure at this virtual re-
ceiver is calculated by averaging the sound pressure at five
points on a plane through this receiver perpendicular to the
canal 共see Fig. 2, inset兲. In the remainder of this paper, every
reference to receiver is a reference to this virtual receiver.
J. Acoust. Soc. Am., Vol. 124, No. 4, October 2008
TABLE I. Correlation coefficients 共cA,B兲 between all pairwise combinations
of the directional component of HRTFs obtained in different locations along
the length of the ear canal, starting at the approximate position of the tym-
panum 共point 1; see Fig. 2兲 and ending at the ear canal entrance 共point 5兲.
Location 1 共tympanum兲 2 3 4 5 共entrance兲
1 1.0000 0.9843 0.9672 0.9596 0.9127
2 0.9843 1.0000 0.9827 0.9646 0.9176
3 0.9672 0.9827 1.0000 0.9820 0.9267
4 0.9596 0.9646 0.9820 1.0000 0.9520
5 0.9127 0.9176 0.9267 0.9520 1.0000
Contrary to the measurements with which the simulation
results are compared, the receiver was not located at the
tympanum itself. This was due to imaging problems: The
tympanum can only be determined approximately on the data
obtained from the microCT machine. Instead, the receiver
was located at the entrance of the ear canal, because, as
shown by several authors,17,29 the directional properties of
the human HRTF are not affected by such a choice. Since the
maximum radius of the ear canal of the bat model has a
diameter of 2.8 mm, which is smaller than the wavelength at
the highest simulated frequency 共3.6 mm兲, we assume that
the dominant propagating modes in the canal will consist of
plane waves as well. Hence, the directional properties of the
bat HRTF are not expected to change in our simulation ei-
ther. The correlation coefficients displayed in Table I do in-
deed validate this assumption by showing that the directional
component of the HRTFs, as defined in Ref. 12, obtained
from different locations in the ear canal 共see Fig. 2兲 are all
very similar. Note that by placing the receivers at the en-
trance of the ear canal, the frequency transfer characteristics
of the ear canal are not modeled. However, those will only
introduce interfrequency amplitude differences, not intrafre-
quency differences and thus only affect the direction-
independent component of the HRTF.
The source positions for which the HRTF is calculated
are located on the frontal hemisphere with center the receiver
location and radius 1 m. The azimuth 共elevation兲 samples for
which data are obtained lay between −90° and 90° 共between
−82.5° and 82.5°兲, spaced at 2.5°. Points at the right of the
midsagittal plane are indicated with negative azimuth values
and points below the horizontal plane with negative eleva-
tion values. The head is tilted so that the plane in which the
nostrils are located is perpendicular to the horizontal plane
共cf. Refs. 2 and 16兲. A schematic of the setup is shown in
Fig. 3.
D. Data processing
Unless stated otherwise, the HRTFs are normalized with
regard to the highest amplitude over the range of frequencies
and simulated source points, not over the simulated source
points per frequency. In this way, the relative strength and
thus the importance of different frequencies is preserved.
Data calculated in the simulations are smoothed by averag-
ing samples in a cube with edge length of five samples, after
which it is converted to decibel values, assuming a noise
floor below −80 dB.
De Mey et al.: Calculating the head related transfer function 2125

FIG. 3. Calculation setup: The bat is oriented so that the plane containing
the nostrils is perpendicular to the horizon; the receiver is located in the
origin of the coordinate system at the entrance of the ear canal. Sound
sources are placed on the frontal hemisphere at a distance of 1 m, spaced at
2.5° for both azimuth and elevation. The head model is not shown to scale
共see Fig. 1兲.
To visualize the HRTF for a specific frequency, Lam-
bert’s azimuthal equal area projection9 is used. As noted by
Müller,24 the use of an equal area projection is desirable
since one can discern the relative cross-sectional area of the
lobes of the beam pattern on those projections. The trade-off
is that angular distortions are introduced and lobe shape will
not be representative. Lambert’s projection, however, keeps
those distortions to a theoretical minimum.24
When comparing different HRTFs, correlation coeffi-
cients are taken between the spatial information per fre-
quency or between the entire frequency–space range. They
are calculated with
兺␪,␸共A␪,␸,f − Ā f 兲共B␪,␸,f − B̄ f 兲
共1兲
cA,B,f =
冑兺␪,␸共A␪,␸,f − Ā f 兲2兺␪,␸共B␪,␸,f − B̄ f 兲2 ,
兺␪,␸,f 共A␪,␸,f − Ā兲共B␪,␸,f − B̄兲
共2兲
cA,B =
冑兺␪,␸,f 共A␪,␸,f − Ā兲2兺␪,␸,f 共B␪,␸,f − B̄兲2 ,
in which A and B are HRTFs, ␪ designates azimuth, ␸ eleva-
tion, and f frequency. Ā f represents the spatial average of the
HRTF A at frequency f and Ā represents the spatial and
frequency average of the HRTF A.
As mentioned earlier, the frequency transfer characteris-
tics of the ear canal are not modeled in simulation. More-
over, they are different for each measured bat. These charac-
teristics result in an unknown scaling of the HRTF patterns at
different frequencies.
Comparing the HRTFs resulting from simulations or
measurements on a per frequency basis, i.e., using cA,B,f , is
not affected by this phenomenon, since this correlation op-
eration is invariant for scaling. Also, since the ear canal fre-
2126 J. Acoust. Soc. Am., Vol. 124, No. 4, October 2008
quency response will be subtracted out by the method to
construct IIDs proposed in Ref. 2 and followed in this paper,
comparing IIDs using cA,B will not be affected by this scaling
either.
However, the ear canal frequency transfer characteristic
does affect the comparison of a measured HRTF with a simu-
lated HRTF or another measured HRTF using cA,B, since the
frequency transfer characteristic of the canal is different for
all those HRTFs. Hence, since HRTFs cannot be compared
directly, the use of the directional component of the HRTF is
proposed instead. In accordance with Ref. 12, this directional
component, i.e., the direction dependent transfer function
共DTF兲, is obtained by subtracting the spatial average of the
HRTF at each frequency. Since, for a logarithmic represen-
tation, the use of DTFs effectively eliminates the canal fre-
quency transfer characteristics, we will be using DTFs when
comparing monaural information of bats with the correlation
measure cA,B.
III. COMPARISON WITH MEASURED DATA
To gain insight in the possibilities versus the limitations
of using simulations to characterize the HRTF of bats, we
compare a simulation of the HRTF of the Phyllostomus dis-
color to measurements done on three bats from the same
species by Firzlaff et al. in Ref. 2.
In Ref. 2 the spatial hearing properties of three different
specimens of the Phyllostomus discolor were characterized
both monaurally 共HRTF兲 as well as binaurally 共IIDs兲.
Throughout Sec. III, the following nomenclature is used: Bat
1, Bat 2, and Bat 3 are the specimens for which the measure-
ment results are given in Ref. 2; Bat BEM is the specimen
used to extract the boundary model on which the simulation
was done.
For HRTFs, the region of highest sensitivity was char-
acterized by three parameters per frequency: elevation and
azimuth of the axis of highest sensitivity and area of the
region of highest sensitivity. Whereas the two first param-
eters locate the region in space, the third gives an indication
of the spatial tuning of the bat’s hearing at a certain fre-
quency. The axis of highest sensitivity per frequency was
extracted by calculating the center of gravity for the −3 dB
contour around the maximum gain 共region of highest sensi-
tivity兲 observed at that frequency. When this region consisted
of two lobes, that lobe containing the maximum gain for that
frequency was considered the region of highest sensitivity.
A. Comparing regions of highest sensitivity
Here, we investigate the extent to which the frequency-
dependent behavior of the calculated region of highest sen-
sitivity differs from that of the measured ones.2
First, it is checked whether the evolution of the axis of
highest sensitivity over frequency for the simulated bat re-
sembles those of the measured bats. To this end, the azi-
muthal and elevation position of this axis is calculated in
accordance with Ref. 2.
As can be seen in Fig. 4共a兲, for the measured specimens
the axis of maximum sensitivity shifts toward the vertical
midplane with increasing frequency. Since the receiver was
De Mey et al.: Calculating the head related transfer function
 60
a renewed increase can be observed. Thus, the region of
Bat 1
45
Bat 2 highest sensitivity displays an elevational frequency scan-
Azimuth in ° Bat 3
Bat BEM ning behavior.
30 The area of the region of highest sensitivity is expressed
by its solid angle as done in Refs. 2 and 9. This is a measure
15
for a two-dimensional angular span in three-dimensional
(a)
0 space in which a complete hemisphere is indicated by 2␲ sr.
30 As depicted in Fig. 4共c兲, all bat regions of highest sensitivity
Bat 1 have a decreasing area, i.e., improved spatial tuning with
15 Bat 2
increasing frequency.
Elevation in °

Bat 3
0 Bat BEM
From these results it is concluded that one of the main
-15 features of the HRTF, i.e., the region of highest sensitivity,
-30
shows very similar behavior for both the calculations and the
(b) measurements.
-45

B. Comparing HRTF patterns

Bat 1
Area in steradians

p/3 Bat 2
Bat 3 Although the region of highest sensitivity already cap-
Bat BEM
tures a significant part of spatial hearing,16 complete HRTF
p/6 patterns still contain considerable extra information.12
Hence, the following analyzes the correspondence between
(c) the complete calculated and measured HRTF patterns.
0
25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 As can be seen from Fig. 5, visualizing the different
Frequency in kHz HRTF patterns again shows good qualitative correspondence
FIG. 4. 共a兲 Azimuth of the axis of highest sensitivity for the measured bats
for all bats and all frequencies. In particular, because in-
共Bat 1, Bat 2, Bat 3; data from Ref. 2兲 and the simulated bat; 共b兲 elevation traspecies variability has to be taken into account, as the
of the axis of highest sensitivity; and 共c兲 area of the region of highest specimen used in the simulation is not one of the three bats
sensitivity, expressed in steradians. used to measure the HRTFs in Ref. 2.
A quantitative comparison of the HRTF patterns is per-
placed in the left pinna, the axis shifts from left to right. The formed using the previously defined correlation measures. To
axis of maximum sensitivity of the simulated HRTF exhibits this end, correlations between all the specimens for which an
the same behavior. HRTF was acquired are calculated. Table II displays the cA,B
Figure 4共b兲 shows the elevation location of the axis of correlations of the HRTFs. From these results it can be seen
highest sensitivity for all the bats. In both the measured that the simulated HRTF denoted by Bat BEM resembles Bat
HRTFs as well as the simulated one, the same trends are 2 and Bat 3 better than the HRTF of Bat 1 does, positioning
displayed: an increase in elevation for the axis of highest the simulated bat well within the population of measured
sensitivity with increasing frequency up to ⬃50 kHz fol- bats.
lowed by a sudden drop around 50– 55 kHz, in which the On a per frequency basis, using the cA,B,f correlation, the
former region of maximum sensitivity splits into two lobes, results depicted in Fig. 6共a兲 are produced. Again, Bat BEM
one of which has a lower sensitivity. For higher frequencies shows good resemblances to the other bats, in the sense that

25 kHz 35 kHz 45 kHz 55 kHz 65 kHz 75 kHz 85 kHz 95 kHz

Bat 1
Bat 2
Bat 3
Bat BEM

FIG. 5. HRTF patterns of the measured bats 共Bat 1, Bat 2, Bat 3; data from Ref. 2兲 and of the simulated bat 共Bat BEM兲 over the entire frequency range.

J. Acoust. Soc. Am., Vol. 124, No. 4, October 2008 De Mey et al.: Calculating the head related transfer function 2127
TABLE II. Correlation coefficients 共cA,B兲 between all pairwise combinations 60 kHz
of HRTFs of the bats.
dB
-50 -40 -30 -20 -10 0 -50 -40 -30 -20 -10 0
Bat 1 Bat 2 Bat 3 Bat BEM
82.5

-3
Bat 1 1.0000 0.7177 0.6443 0.6701

3
55.0 -24

-30
Bat 2 0.7177 1.0000 0.8083 0.8114 -36 -21

Elevation in °
27.5 -33 -27 -1
Bat 3 0.6443 0.8083 1.0000 0.7719 -30
-21
-15 8
-24
Bat BEM 0.6701 0.8114 0.7719 1.0000 -27

-12
0

-30
-18

-9
-30 -12
-27.5

5
-1

-27
-6

-21
-24 4
-2
-55.0 -27
it differs from the other bats as much as the other bats differ (a) -15
(b)
-21

from each other. The only outlier in the correlation data oc- -82.5

curs at 60 kHz: The correlation between Bat 1 and the other

dB
ones is especially bad there. Figure 7 displays the HRTF -50 -40 -30 -20 -10 0 -50 -40 -30 -20 -10 0
pattern for the bats at this frequency. Despite the quantitative 82.5 -2
-2 7
differences between all four bats, we clearly see that Bat 2, 7
55.0 -27
Bat 3, and Bat BEM display the same qualitative HRTF char-

-30
-21 -24
-18

Elevation in °
-21 -3
27.5

0
acteristics: The region of highest sensitivity has comparable 3 -15

-3
- 30
shape and is located in the lower left quadrant 共azimuth:

-24
-6
0 -2

-1
-1 4 -3

2
2
0–90°; elevation 0 to − 90°兲. Notches are located in the -2
7
-6 -9

-2
4
-27.5
perifery. Bat 1, however, has a less pronounced peak. More- -9
-1
5
-30

-55.0 -18
over, it has additional sidelobe features, especially at higher -27
-27
(c) (d) -24
elevation values. -82.5
-90 -60 -30 0 30 60 90 -90 -60 -30 0 30 60 90
From those correlation values as well as from the dis- Azimuth in ° Azimuth in °
played HRTFs, it is again concluded that the simulated
HRTF is that of a bat among bats, i.e., no significant differ- FIG. 7. HRTFs of the four bats at 60 kHz: 共a兲 Bat 1; 共b兲 Bat 2; 共c兲 Bat 3; and
ences between the simulated and the measured HRTFs are 共d兲 Bat BEM.
apparent. However, despite the good overall correspondence
between the simulations and the measurements, as shown in
Fig. 5, some discrepancies remain. In particular, the dynamic sorption will determine the strength of the contralateral
range of the HRTF patterns for Bats 1, 2, and 3 共⬃40 dB兲 is sound. The rigid boundary assumption affects the sound
considerably larger than that of Bat BEM 共⬃24 dB兲. We coming from the ipsilateral side much less, as these sound
conjecture that this is one instance where the absence of waves are directly captured by the pinna for which this as-
realistic impedance information for the boundary affects the sumption is much more appropriate.
results. Indeed, noting that sound coming from the contralat-
eral side is diffracted around the head before it enters the
ipsilateral ear, it can be assumed that it will also be absorbed
by the fur on the bat’s head. For the simulated bat, though, C. Comparing IID patterns
with its acoustically rigid boundary, diffraction without ab-
Most experimental approaches exploit the gross symme-
try of the bat’s head to obtain IID patterns: The measure-
1.0
ments of the monaural HRTF are mirrored over the vertical
Correlation Coefficient

0.8 midplane to obtain the response of the contralateral ear and

0.6 then they are subtracted from each other.2,4,16 Such an ap-
proach simplifies the experimental setup considerably, espe-
0.4
cially for small animals like bats, as it allows one to derive
0.2 Bat 1 - Bat 2 Bat 1 - Bat BEM Bat 2 - Bat BEM binaural information from monaural microphone measure-
(a) Bat 1 - Bat 3 Bat 2 - Bat 3 Bat 3 - Bat BEM
0 ments. For a simulation approach like the one proposed in
1.0 this paper, such mirroring does not provide any advantages,
because calculating the HRTF for both ears simultaneously is
Correlation Coefficient

0.8
just a matter of specifying more receivers, resulting in a neg-
0.6
ligible extra computational load. However, in this paper, this
0.4 multiple receiver capability is not taken advantage of, as the
0.2 Bat 1 - Bat 2 Bat 1 - Bat BEM Bat 2 - Bat BEM same symmetry property is employed to make the results
(b) Bat 1 - Bat 3 Bat 2 - Bat 3 Bat 3 - Bat BEM directly comparable with the measurements.2
0
25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 Table III and Fig. 6共b兲 show similar results as in Sec.
Frequency in kHz III B: Bat BEM resembles Bat 2 and Bat 3 better than Bat 1
does. So, as with the previous two comparison criteria, bin-
FIG. 6. 共a兲 Correlation coefficients 共cA,B,f 兲 between all pairwise combina-
tions of the HRTFs of the bats; and 共b兲 correlation coefficients 共cA,B,f 兲 be- aural IID cues also indicate that the simulated bat behaves
tween all pairwise combinations of the IIDs of the bats. quite similarly to the measured ones.

2128 J. Acoust. Soc. Am., Vol. 124, No. 4, October 2008 De Mey et al.: Calculating the head related transfer function
TABLE III. Correlation coefficients 共cA,B兲 between all pairwise combina- their functional relevance for the bat’s echolocation. In Ref.
tions of IIDs of the bats. 25 a similar study was performed to analyze the contribu-
Bat 1 Bat 2 Bat 3 Bat BEM
tions of pinna features. In analogy, the head of Phyllostomus
discolor is considered to consist of various acoustically con-
Bat 1 1.0000 0.7647 0.7555 0.7515 spicuous subcomponents: the head, the ipsilateral and con-
Bat 2 0.7647 1.0000 0.7521 0.8249 tralateral pinna, and the noseleaf.
Bat 3 0.7555 0.7521 1.0000 0.7631
Bat BEM 0.7515 0.8249 0.7631 1.0000
A. Influence of the subcomponents on the HRTF
The results shown in Fig. 8, i.e., the differences between
IV. INFLUENCE OF DIFFERENT PARTS OF THE HEAD
the HRTF of a complete head and the HRTFs of the same
Here we discuss an important application that is made head with various subcomponents missing, give a global im-
possible by having access to simulation data: the assessment pression of how the removal of the different subcomponents
of contributions made by various head structures to the of the complete head affects the calculated HRTFs. First, it is
HRTF and the IID. noted that the differences between the HRTF of the complete
By using a virtual model instead of a real head, the head and the HRTFs with missing noseleaf and missing con-
morphology can be easily altered. Morphology changes that tralateral pinna are very small, indicating that those two sub-
would involve irreversible surgery on a real head can be components do not seem to interact much with the acoustic
reversed in virtual reality, making possible the systematic field around the head. On the other hand, the differences
study of the contributions made by various head parts and between the HRTF of the complete head and the HRTF of

Pinna Contralateral Pinna Noseleaf Removed

Removed
90 20
18
60
16
14
Elevation in °

30
12
35 kHz

0 10
8
-30 6
8 4 4
-60 10
6
4 2
-90 0
dB
90 86
20
4

18
6

60 8 4
4

16
4
46

14
Elevation in °

30
4
4
4

12
60 kHz

0 10
4

8
4
6

-30 64 4
6
2

4
6
4

6 4
-60
4

8 4 2
0 8
8 46 1 4
6

-90 0
dB
90 8 20
4
12

4
18
12
6

60 8 6
4
4 16
8

14
Elevation in °

30 10 4
8

12
75 kHz

14
6

0 10
4

12
8
6

4 8
10 12

8
-30 6
8

6
4

8
4

4
4

4
-60 18 16 4
12

14 2
10

8 1102 4
10

-90 0
-90 -60 -30 0 30 60 90 -90 -60 -30 0 30 60 90 -90 -60 -30 0 30 60 90
Azimuth in ° Azimuth in ° Azimuth in °

FIG. 8. Differences between the HRTF patterns of a complete head and, from left to right, the ipsilateral pinna only, the head without the contralateral pinna,
and the head without the noseleaf.

J. Acoust. Soc. Am., Vol. 124, No. 4, October 2008 De Mey et al.: Calculating the head related transfer function 2129
 1.0 1.0

Correlation Coefficient

Correlation Coeffcient
0.8 0.8

0.6 0.6

0.4 0.4
Frontal Hemisphere (FH)
0.2 Contralateral part of the FH 0.2 Ipsilateral pinna removed
Ipsilateral part of the FH Ipsilateral pinna only
0 0
25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95
Frequency in kHz Frequency in kHz

FIG. 9. Correlation coefficients 共cA,B,f 兲 between the HRTF generated by the FIG. 10. Correlation coefficients 共cA,B,f 兲 between the IIDs generated by a
ipsilateral pinna only and the HRTF generated by the complete head, respec- complete head, and respectively, a head from which the ipsilateral pinna is
tively, for the frontal hemisphere, its ipsilateral side, and its contralateral removed and a pinna only.
side.

the pinna only model, while very small at low frequencies, ⬎ 0.8兲, providing evidence in favor of the pinna only ap-
grow larger at higher frequencies. In addition, it can be seen proach employed in Refs. 24 and 25. However, splitting up
that the discrepancies between these two models are biased the global correlation into separate correlation measures for
toward the contralateral half of the frontal hemisphere. the ipsilateral and the contralateral part of the frontal hemi-
The very minor effect on the HRTF of the presence of
sphere reveals a consistently high ipsilateral correlation but a
the contralateral pinna as well as the noseleaf can be quan-
considerably lower contralateral correlation from 55 kHz on,
tified by calculating the correlations between the HRTF of
the complete head and the HRTFs after removal of the con- indicating, as mentioned before, that the complete head and
tralateral pinna 共cA,B,f ⬎ 0.9兲 and removal of the noseleaf the pinna only models differ most in predicting the contralat-
共cA,B,f ⬎ 0.98兲. Similarly, the good correspondence between eral part of the HRTF for the higher frequencies.
the HRTF of the complete head and the HRTF of the pinna This result can be understood by noting that the effect of
共positioned as if it were still attached to the head兲 can also be the head itself is expected to be most pronounced for this
quantified 共see Fig. 9兲 by calculating the correlation 共cA,B,f case, i.e., contralateral sound at high frequencies.

Pinna Removed Head Pinna Only

90 30
3
60
12

20
9
6
-15

-9

15

-6 -3
3

9
Elevation in °

30 -9
-12
-15
12
10
1518 15
0

18
36 kHz

15
-18

-21 21
9
0

0 24 0
24

18

-1
-21
-21

-3

21

5
18

-21
0

-18 -6 21
18
18 21

-15

6 -10
-30
-12
-21

-1 27
8
-60 -12
-20

-90 -30
dB
90 30
0

6 3 -3
3 -9
60 0 9
-9

0 0 20
-3

-3 3 3
3
12
-12

12
12

-12
Elevation in °

30 -9 -6 6 10
3
-3

9
3
76 kHz

-1
21
0

-12
12
8
6 18
9
-9

0 -1 0
-6

-1
0

-6 -1
-3
-18

5 2 12 15 18
-3

5
15
0

-15
-9

0
9

-9
-3
-12

3
-30 -3 0 3
12 -10
-1
0
0
3

2 -18

9
0

-3 -6
3

3 3 18 -20
-60
-3

0
-15

-3
6

-12
15

-6 12
-3

3 6
-90 -30
-90 -60 -30 0 30 60 90 -90 -60 -30 0 30 60 90 -90 -60 -30 0 30 60 90
Azimuth in ° Azimuth in ° Azimuth in °

FIG. 11. IID patterns for the depicted models at 36 and 76 kHz.

2130 J. Acoust. Soc. Am., Vol. 124, No. 4, October 2008 De Mey et al.: Calculating the head related transfer function
 36 kHz 76 kHz
the head afterwards. In this way, correct position and orien-
dB
-30 -20 -10 0 10 20 30 -30 -20 -10 0 10 20 30
tation of the pinna with respect to the head is guaranteed.
80.0 3 -6
Figure 12 indicates that errors in orientation of the pinna will
distort the IID patterns. In particular, azimuth errors will

9
55.0 -9 9

-9
6
-9 9 have the biggest effect on the IID pattern as they result in

6
-9 3
Elevation in °

12 9
27.5 -6

18
9
15

0
-18 15
18 nonrigid deformations, whereas elevation errors will, to a

-3
Pinna

21
18

-15
24
0 first approximation, result in a rigid shift of the IID pattern

-18
21 -12 12
-21

18
18
-9-6
-18

over the elevation offset.

12
-12
-27.5

3
9
24
-1
-55.0 8

21
12
-80.0 V. CONCLUSION
-6

80.0
This paper calculates the HRTF and the IID patterns on
3 -6
55.0
6
-6 6
3 a model of the complete head of a Phyllostomus discolor,
6
-3
0 with a fast BEM tool designed for HRTF calculation. The
Az. error = 10°

-6
Elevation in °

27.5 -9 3
9

6
tools to acquire and to process the 3D model, yielding the

-6
-15- 12 15 6 -3
9

18
18
2

12
-6
0 -1
HRTF, have only recently become available. The obtained

6
15

-3

9
-12 3
9
results are in good agreement with acoustic measurements on
-6

-27.5 0

9
18

0 3
9
specimens of the same species. In addition, it is noted that
21

-9 0
-55.0 -15
15

15
-80.0
the remaining discrepancies between the model and the mea-
surements are most pronounced in the contralateral half of
80.0
-6 9
the frontal hemisphere. It is conjectured that this part of the
0 3 6 -3
55.0 0 sound field is determined to a larger extent by diffraction
6

-6 0 0
-
Az., El. error = 10°

3 -9 3
around and absorption on 共fur兲 the bat’s head, and hence, that
-9 9 6
0
0

-12 9
Elevation in °

27.5 15 -1
-6

5
-1-1 12 5 -6 6
15

8
6

this is one area where knowledge of realistic impedance in-

12
2
-1

18
0
-9

18
9

-9 formation would allow further improvement of the results.

93
-27.5 -3 - 0
0

Once validated by comparison with real measurements,

15

6 0
0
-15

12
15

-15
-55.0 15
this simulation approach is shown to have a large potential
-12

-3

-80.0
-90 -60 -30 0 30
Azimuth in °
FIG. 12. IID patterns at 36 and 76 kHz for the pinna only model. Correct
pinna orientation 共top row兲; pinna orientation with azimuth error 10°
共middle row兲; and pinna orientation with both azimuth and elevation error
10° 共bottom row兲.
B. Influence of the subcomponents on the IID
Finally, the shadowing effect of the head on the IID
pattern is appraised by cutting away the ipsilateral pinna in
one model and comparing the resulting IID cues to those
corresponding with a complete model and a pinna only
model.
Inferring from the results in Figs. 10 and 11, it is con-
cluded that the contribution of the ipsilateral pinna is a con-
siderably more important factor in shaping the IID than the
head itself is. Indeed, for all frequencies the IID patterns for
the pinna only model resemble the complete head model
much better than do the IID patterns for the head without
pinna model, in accordance with the results in Sec. IV A,
which already indicated the importance of the ipsilateral
pinna for the HRTF.
As was the case for the HRTF comparison, the results in
Fig. 11 also show that, while for lower frequencies the pinna
only model and the complete head model result in very simi-
lar predictions for the IID patterns, discrepancies appear at
the higher frequencies.
It should be noted that when replacing the complete
head by a pinna only model, it is still advantageous to start
from a model of the entire bat head and virtually cut away
-9 9
for analyzing “what-if” scenarios. In particular, simulations
60 90 -90 -60 -30 0 30 60 90
Azimuth in °
are used in this paper to determine the effect of the different
subcomponents of the bat’s head on the resulting HRTF and
IID patterns. It is concluded from this analysis that, for this
species, the ipsilateral pinna is the single most important
subcomponent of the head. It is also shown that if the com-
plete head model is replaced by a pinna only model for simu-
lation, precise positioning and orienting of this pinna is es-
sential.
The application of the simulation method paves the way
for further experiments and research, e.g., the study of the
interplay between the directivity of the noseleaf and the
HRTF, i.e., sender–receiver interaction, and the calculation
of the HRTF for dynamically deforming pinnae. So far,
simulation studies have assumed the pinnae-head configura-
tion to be fixed. Several species, however, actively position/
deform their pinnae during the echolocation process.
ACKNOWLEDGMENTS
The authors would like to thank N. De Clerck, A. Post-
nov, and F. Lakiere, from the MCT group at the Universiteit
Antwerpen and S. Ise at Kyoto University for his contribu-
tions to the BEM3D software used in this paper. This work
was funded through the CILIA project 共EU-FET兲.
1
E. Young, J. Rice, and S. Tong, “Effects of pinna position on head-related
transfer functions in the cat,” J. Acoust. Soc. Am. 99, 3064–3076 共1996兲.
2
U. Firzlaff and G. Schuller, “Spectral directionality of the external ear of
the lesser spear-nosed bat, phyllostomus discolor,” Hear. Res. 181, 27–39
共2003兲.
3
B. Lawrence and J. Simmons, “Echolocation in bats: The external ear and
perception of the vertical positions of targets,” Science 218, 481–483
共1982兲.
4
T. Shimozawa, N. Suga, P. Hendler, and S. Scheutze, “Directional sensi-

J. Acoust. Soc. Am., Vol. 124, No. 4, October 2008 De Mey et al.: Calculating the head related transfer function 2131

tivity of echolocation system in bats producing frequency-modulated sig-
nals,” J. Exp. Biol. 60, 53–69 共1974兲.
5
J. Wotton, T. Haresign, and J. Simmons, “Spatially dependent acoustic
cues generated by the external ear of the big brown bat, eptesicus fuscus,”
J. Acoust. Soc. Am. 98, 1423–1445 共1995兲.
6
J. Wotton and J. Simmons, “Spectral cues and perception of the vertical
position of targets by the big brown bat, eptesicus fuscus,” J. Acoust. Soc.
Am. 107, 1034–1041 共2000兲.
7
R. Butler and K. Belendiuk, “Spectral cues utilized in the localization of
sound in the median sagital plane,” J. Acoust. Soc. Am. 61, 1264–1269
共1977兲.
8
R. Humanski and R. Butler, “The contribution of the near ear and far ear
toward localization of sound in the sagittal plane,” J. Acoust. Soc. Am. 83,
2300–2310 共1988兲.
9
S. Carlile, “The auditory periphery of the ferret. I. Directional response
properties and the pattern of interaural level differences,” J. Acoust. Soc.
Am. 88, 2180–2195 共1990兲.
10
A. Musicant, J. Chan, and J. Hind, “Direction-dependent spectral proper-
ties of cat external ear: New data and cross species comparisons,” J.
Acoust. Soc. Am. 87, 757–781 共1990兲.
11
L. Xu and J. Middlebrooks, “Individual differences in external-ear transfer
functions of cats,” J. Acoust. Soc. Am. 107, 1451–1459 共2000兲.
12
M. Aytekin, E. Grassi, M. Sahota, and C. Moss, “The bat head-related
transfer function reveals binaural cues for sound localization in azimuth
and elevation,” J. Acoust. Soc. Am. 116, 3594–3605 共2004兲.
13
T. Park, B. Grothe, G. Pollak, G. Schuller, and U. Koch, “Neural delays
shape selectivity to interaural intensity differences in the lateral superior
olive,” J. Neurosci. 16, 6554–6566 共1996兲.
14
J. Blauert, Spatial Hearing 共MIT, Cambridge, 1997兲.
15
M. Brainard, E. Knudsen, and S. Esterly, “Neural derivation of sound
source location: Resolution of spatial ambiguities in binaural cues,” J.
Acoust. Soc. Am. 91, 1015–1027 共1992兲.
16
M. Obrist, M. Fenton, J. Eger, and P. Schegel, “What ears do for bats: A
comparative study of pinna sound pressure transformation in chiroptera,”
J. Exp. Biol. 180, 119–152 共1993兲.
2132 J. Acoust. Soc. Am., Vol. 124, No. 4, October 2008
17
J. Middlebrooks, J. Makous, and D. Green, “Directional sensitivity of
sound-pressure levels in the human ear canal,” J. Acoust. Soc. Am. 86,
89–108 共1989兲.
18
B. Katz, “Boundary element method calculation of individual head-related
transfer function. I. rigid model calculation,” J. Acoust. Soc. Am. 110,
2240–2448 共2001兲.
19
B. Katz, “Boundary element method calculation of individual head-related
transfer function. II. Impedance effects and comparison to real measure-
ments,” J. Acoust. Soc. Am. 110, 2449–2455 共2001兲.
20
M. Otani and S. Ise, “A fast calculation method of the head-related trans-
fer functions for multiple source points based on the boundary element
method,” Acoust. Sci. & Tech. 24, 259–266 共2003兲.
21
M. Otani and S. Ise, “Fast calculation system specialized for head-related
transfer function based on boundary element method,” J. Acoust. Soc. Am.
119, 2589–2598 共2006兲.
22
T. Walsh, L. Demkowicz, and R. Charles, “Boundary element modeling of
the external human auditory system,” J. Acoust. Soc. Am. 115, 1033–1043
共2004兲.
23
T. Xiao and Q. Liu, “Finite difference computation of head-related transfer
function for human hearing,” J. Acoust. Soc. Am. 113, 2434–2441 共2003兲.
24
R. Müller, “A numerical study of the role of the tragus in the big brown
bat,” J. Acoust. Soc. Am. 116, 3701–3712 共2004兲.
25
R. Müller, H. Lu, S. Zhang, and H. Peremans, “A helical scanning pattern
in the Chinese noctule, nyctalus plancyi,” J. Acoust. Soc. Am. 119, 4083–
4092 共2006兲.
26
Y. Naka, A. Oberai, and B. Shinn-Cunningham, “The finite element
method with the Dirichlet-to-Neumann map for sound-hard rectangular
rooms,” in Proceedings of the 18th International Congress on Acoustics,
Kyoto, Japan, April 2004.
27
http://www.skyscan.be/products/1076.html; last viewed 24 January 2008.
28
K. H. Esser and A. Daucher, “Hearing in the fm-bat phyllostomus dis-
color: A behavioral audiogram,” J. Comp. Physiol., A 178, 779–785
共1996兲.
29
D. Hammershoi and H. Moller, “Sound transmission to and within the
human ear canal,” J. Acoust. Soc. Am. 100, 408–427 共1996兲.
De Mey et al.: Calculating the head related transfer function