Exterior Auralization of Traffic Noise within the LISTEN project

Andrew Peplow
Hoare Lea Acoustics, 140 Aztec West, Bristol, BS32 4TX, UK

Jens Forssn
Applied Acoustics, Chalmers University of Technology, SE-412 96 Gteborg, Sweden

Peter Lundn
Interactive Institute, SoundSpace studio, SE-164 26 Kista, Sweden

Mats E. Nilsson
Institute of Environmental Medicine, Karolinska Institutet and Department of Psychology,
Stockholm University, SE-106 91, Stockholm, Sweden
Summary
In the present paper, computational auralizations of a single pass-by of a passenger car and
passenger train are investigated. Auralization of future traffic noise scenarios would be a valuable
tool for city planners, noise consultants and decision makers, since it would make it possible to
evaluate various noise mitigation solutions already at the planning stage. The main goal of the
Swedish multidisciplinary research project LISTEN has been to develop such a tool. This paper
reports recent listening experiments conducted to perceptually evaluate the auralizations and
subsequent improvements on computational methodologies regarding road and rail traffic noise and
passenger train source models.
PACS no. 43.20.Ei, 43.66.Lj

1.

Introduction1

Traffic noise is a major environmental problem,

affecting the health and wellbeing of a large part of
the population. Within the Swedish multidisciplinary research project LISTEN, we are
developing a demonstrating tool for noise and noise
abatement of road and rail traffic, which is aimed at
helping assessment processes at various levels of
decision making. The project started in September
2008. The work presented here shows results of
auralization of single pass-bys of passenger cars
and passenger trains. The auralization methodology
is mainly based on a time-domain approach.
Psychoacoustic evaluation was conducted on all car
simulations
concerning
perceived
realism,
annoyance and speed.

2.

Methodology for auralization

In the processes of auralizing road traffic noise, for

a still-standing person at a position exterior to the
vehicle, a number of parameters need to be
considered. The path of the sound that reaches the
listener from a single vehicle can be described as
starting from a set of sources that each has its own
property concerning directivity and spectral
content, including noisy and tonal characteristics.
1

(c) European Acoustics Association

For the passenger car modelled here, two sources

are used, (i) a propulsion source, which models
engine, air intake, air exhaust, fans, compressors
(traction), etc., and (ii) a road/tyre source, which
models the noise generated by the contact between
tyre and road surface. This set is modelled
according to current engineering methods
(Harmonoise and Nord2000 [1,2]), with the sources
located on a vertical line at heights (i) 0.3 m and (ii)
0.01 m. For the auralization, the source signals are
modelled as steady-state noises and steady-state
tones. After leaving the source, the sound on its
path will be influenced by: decaying amplitude due
to spherical spreading; air attenuation, which leads
to a larger reduction of the higher frequencies than
of the lower frequencies; and the reflection in the
ground surface, which leads to an interference
pattern over frequency. The theoretical interference
pattern for a point source can show very deep dips.
The real interference pattern is however usually
weakened due to smearing effects like decorrelation
caused by random ground roughness and air
turbulence as well as effects of the finite sized
sources, which has been considered here. For the
moving source, relative to the receiver, we also
have the Doppler effect. Here, we have not
considered the effects of refraction, i.e. the curved
sound paths due to wind or height varying
temperature, which however will be influential at
longer distances. Finally, the sound entering our
ears is influenced by our head and torso, which here

FORUM ACUSTICUM 2011

27. June - 1. July, Aalborg

Forssn, Exterior auralization of traffic noise within the LISTEN project

is modelled using head related transfer functions

(HRTFs) from the CIPIC database, subject 165, a
KEMAR manikin [3]. The whole auralization
process can in our modelling be summarized as
starting with the creation of the noises and tones of
the sources followed by the modelling of the air
attenuation, the ground effect, the directivity, the
Doppler effect, the spherical spreading and ending
with the HRTFs. Modelling in 1/3 octave bands is
undertaken for the noise signals, the air attenuation,
the ground effect and the directivity, whereas the
remaining effects are modelled directly on the time
signal. The creation of the source signals is based
on mono recordings of straight-line pass-bys at
fixed points (6 second long recordings). The
recorded signals are then reshaped into to source
signals by applying the knowledge of the sound
propagation as an inverse problem. Similar
approaches have been published previously [4,5].
Following the above description of the
methodology, the process of going from the mono
recording to the source signal can be described as
inverting the effect of spherical spreading, the
Doppler effect, the ground effect, and the airattenuation. The signal resulting from the inversion
process is seen as a first estimate of the source
signal.

midrange position.) The tonal characteristics are

modelled from peaks of a narrow band spectrum of
the source signal, here using one tone per 1/3
octave band. The modelling of the propagation
effects within the auralization process uses the
same tools as the inversion process for the estimate
of the source character, as described above.
value used here for a dense asphalt surface is 2108
Nsm-4. From that, the ground impedance is
calculated (e.g. [1]), which enables calculation of
the spherical reflection factor. The remaining
distance effect is according to spherical spreading.
Finally, the Doppler effect is calculated simply
using a resampling of the signal or, equivalently,
seen as a delay line.
For the train auralizations a directivity factor has
been included in the methodology according to the
work by Bongini and Bonnet [6] for the traction
noise and the rolling noise. The switch from
monopole to a dipole-like source was implemented
for rolling noise whereas for traction sources which
dominate low-frequency ranges any modifications
to directivity did not influence the global pass-by
signature.

3.
The estimated source signal resulting from the
inversion process described above is seen to contain
a steady-state signal shaped by the source
directivity. The slowly varying amplitudes
(envelopes) of each 1/3 octave band are hence
referred to as source directivity.
For cars the pass-by pattern (in dB) is separated
into forward and backward direction, to which
second-degree polynomials are fitted. At midrange
between forward and backward direction, i.e.
perpendicular to the driving direction, the two
polynomials are adjusted to give a continuous
curve. In this way the polynomial coefficients for
each 1/3 octave band are estimated and saved for
later use in the auralization.
The transition frequency of source dominance is
estimated from the Harmonoise source model [1].
The remaining signal, after compensating for the
estimated directivity patterns, is in most cases a
fairly steady signal. However, at midrange position
transient characteristics are still present, not
captured by the polynomial fitting. Therefore, a
period shortly after passage at midrange position is
used as an estimate of the source signal. (Here, a 1 s
long signal is used, starting 1 s after passage at

Perceived
annoyance

realism,

speed

and

In the first listening experiment of car pass-bys,

pairs of sounds were presented, each pair consisting
of one real recording and one auralization of the
same recording. The task was to decide which of
the two sounds that was the real recording. If a
listener clearly could hear which sound was real
and which was auralized, then the proportion of
correct responses, p(c), would be close to 1.0. If a
listener could not discriminate between the two
sounds, the expected p(c) would be 0.5. It is also
possible that a listener could discriminate between
the two sounds, but could not identify which was
real and which was auralized. Such a listener would
have a p(c) close to 0.0 (mainly incorrect choices)
or a p(c) close to 1.0 (mainly correct choices).
Figure 2 illustrates these results for a car pass-by.
Further details of the results may be found in [7].
The listening panel also assessed which of two car
passages, a real recording and its auralization, that
was more (a) annoying or (b) was perceived as
having the higher speed. For the auralized sounds
these were assessed as more annoying than the real
sounds in about 50 % of the cases, suggesting that
auralized and real sounds were approximately

FORUM ACUSTICUM 2011

27. June - 1. July, Aalborg

Forssn, Exterior auralization of traffic noise within the LISTEN project

distances and the three segments of the car passage.

Chance performance corresponds to 50 % correct
responses, perfect performance to 100 % correct
responses.

4.

Figure 2 : Proportion of responses correctly identifying

which of two sounds corresponds to real recording.

equally annoying. The pattern of results for

perceived speed was found to be similar as for
perceived annoyance.
110

Train auralisation

In adopting the synthesis method described in

section 2 one has to define the rolling stock,
traction and track as a set of noise point sources.
The synthesis method determines each noise source
pass-by into a time-frequency environment, taking
into account, its time-evolution (depending on passby scenario), the Doppler effect and the ground
effect. The time-frequency plans, each of them
corresponding to an equivalent point source
contribution towards the pass-by noise, are
summed. The global time-frequency plan is
therefore transposed into the time domain. The
main consequence of the synthesis method on the
source model, is that the point source model is
considered for equivalent source. For passenger car
realizations this adoption of point source models
seems reasonable upon listening panel analysis.
However, for train auralizations it is still unclear
whether a global point-source methodology is
viable for all passenger train types.

100

It must be pointed out that most of trains in urban

Percent correct responses

90
80
70
60
50
40
30

Part of sound
Start

20

Middle
10
0

End
30

50

70

90

110

Speed (km/h)

Figure 3. Result from a same-different experiment

Figure 3 illustrates a test where listeners had to

decide whether two sounds were identical or
different. The sounds were 2-s segment from the
start, middle or end part of a 6-s car passage. The
sound pairs consisted of either two identical (both
real recordings or both auralized ) or two different
sounds (one real and one auralized). The average
percent of correct responses (+/- 1 standard error)
from 10 listeners are shown separately for the five

Figure 4: Illustration showing location of equivalent

point sources in the train model for IC6 inter-city train.

rolling conditions are composed of about twenty

real noise sources. Moreover, since this method has
been developed to carry out parametric studies on
the influence of each noise source on the global
pass-by noise, the equivalent sources have to be
easily handled according to the modifications
managed on real sources. Consequently, it is

FORUM ACUSTICUM 2011

27. June - 1. July, Aalborg

Forssn, Exterior auralization of traffic noise within the LISTEN project

recommended to represent one physical noise

source by very few point sources preferably only
one. An RC6 intercity train shown in Figure 4 was
modeled by 17 point sources, whereas a smaller
commuter train Regina was modelled according to
a seven equivalent point source scheme.
Measurements undertaken by WSP Sweden
correspond to rail measurements adopted in the ISO
3095 standard for the pass-by measurement. This
method consists on recording pass-by noise with a
microphone located at 7.5m from the track. It
allows a characterisation of an isolated source level
spectrum. However, as soon as several sources are
close together, the contribution of each physical
source cannot be separated. Since it does not locate
equivalent point source it does not provide
directivity patterns. These short-comings are
partially resolved in the auralization dedopplerisation methodology. Results showing time
signals for an auralization of an original recording
and a scenario including a small screen at track-side
are presented in the following figures. The train
length is 55.0m and estimated to travel at 90 km/h.
The noise level at the receiver is estimated at
LAFmax=95 dB(A).

Figure 6: Time signal auralization for Regina commuter

train corresponding to Figure 4.

A small rigid noise screen 0.5m high and 1.5m

from the track-side was included in the
auralization incorporating the Hadden-Pierce
diffraction model [8]. A reduced noise level
compared to Figure 6 calculated at 7.5 m from the
source, around LAFmax=85 dB(A), is clearly
displayed in Figure 7. Although not visible here,
the high frequency squeal noise has been
attenuated significantly.

Figure 7: Time signal auralization for Regina commuter

train including a small noise screen.

5.
Figure 5: Time signal recording for Regina commuter
train

Figure 5 shows the time signal derived from a

Regina pass-by recording. Inspection of the figure
showing the onset of a pulse-type wave, as the
first wagon passes by the microphone, is clearly
evident. This is an aerodynamic effect, and
evidence of the difficulty in modelling this as a
source may be seen in Figure 6, where an
auralization for this situation has been performed.

Conclusions

The results for passenger cars showed a useful

auralization methodology. Although listeners could
identify which of two sounds was the real recording
more often than expected by chance, they were
never 100 percent accurate, and in several cases
close to chance level, especially for higher speeds.
For the auralized sounds from passenger cars,
further listening test results will be presented in the
technical sessions. Results for train auralizations
are promising, technical results and auralizations
will also be presented.
Acknowledgement
This research was conducted within the Swedish
research program LISTEN. Auralization of Urban
Soundscapes funded by The Knowledge
Foundation, Invest in Sweden Agency, Swedish
Governmental Agency for Innovation Systems

FORUM ACUSTICUM 2011

27. June - 1. July, Aalborg

Forssn, Exterior auralization of traffic noise within the LISTEN project

(VINNOVA), Swedish Foundation for Strategic

Research, and The Vrdal Foundation. Co-funding
was provided by the LISTEN projects end-user
partners: The Swedish Road Administration,
Swedish Rail Administration, Stockholm City,
WSP Acoustics, F-Ingemansson, and Rambll
Denmark A/S.
References
[1.] Nota, R., Barelds, R. and van Maercke, D.
Engineering method for road traffic and railway
noise after validation and fine-tuning. Technical
Report HAR32TR-040922-DGMR20 2005.
[2.] Jonasson, H. Acoustic Source Modelling of Nordic
Road Vehicles. SP Rapport 2006:12, 2006.
[3.] Algazi, V. R., Duda, R. O., Thompson, D. M.,
Avendano, C. The CIPIC HRTF Database. Proc.
2001 IEEE Workshop on Applications of Signal
Processing to Audio and Electroacoustics, 99-102.
[4.] Maillard, J., Martin, J. A simulation and restitution
technique for the perceptive evaluation of road
traffic noise. Proc. 2008 Euronoise, Paris.
[5.] T. Kaczmarek. Road-vehicle simulation for
psychoacoustic studies. Proc. 2007 ICA, Madrid.
[6.] E. Bongini and E. Bonnet. Railway noise sources
definition within the scope of pass-by sound
synthesis. Proc 2009, Euronoise, Edinburgh.
[7.] J. Forssen, T. Kaczmarek, J. Alvarsson, P. Lunden,
M.E. Nilsson, Auralization of traffic noise within
the LISTEN project Preliminary results for
passenger car pass-by.
[8.] A. D. Pierce. Diffraction of sound around corners
and over wide barriers. J. Acoust. Soc. Am. 97(4)
(1995) 2028-2040.