Applied Acoustics 117 (2017) 219–226

Contents lists available at ScienceDirect

Applied Acoustics
journal homepage: www.elsevier.com/locate/apacoust

A study of the accuracy of mobile technology for measuring urban noise

pollution in large scale participatory sensing campaigns
Pierre Aumond a, Catherine Lavandier a,⇑, Carlos Ribeiro b, Elisa Gonzalez Boix c, Kennedy Kambona c,
Ellie D’Hondt c, Pauline Delaitre a
a
Laboratoire Mobilité, Réseaux, Territoires et Environnement, Université de Cergy Pontoise, France
b
Bruitparif, France
c
Software Languages Lab, Vrije Universiteit Brussel, Belgium

a r t i c l e i n f o a b s t r a c t

Article history: The study reports on the relevancy and accuracy of using mobile phones in participatory noise pollution
Received 14 December 2015 monitoring studies in an urban context. During one year, 60 participants used the same smartphone
Received in revised form 13 July 2016 model to measure environmental noise at 28 different locations in Paris. All measurements were per-
Accepted 14 July 2016
formed with the same calibrated application. The sound pressure level was recorded from the micro-
Available online 26 July 2016
phone every second during a 10-min period. The participants frequently measured the evolution of the
sound level near two standard monitoring sound stations (in a square and near a boulevard), which
Keywords:
enables the assessment of the accuracy and relevancy of collected acoustic measurements. The instanta-
Sound level monitoring
Mobile phones
neous A-weighting sound level, energy indicators such as LA,eq, LA10, LA50 or LA90 and event indicators such
Crowdsourcing as the number of noise events exceeding a certain threshold La (NNEL P La) were measured and com-
Participatory sensing pared with reference measurements. The results show that instantaneous sound levels measured with
Urban studies mobile phones correlate very well (r > 0.9, p < 0.05) with sound levels measured with a class 1 reference
sound level meter with a root mean square error smaller than 3 dB(A). About 10% of the measurements
for the boulevard location (respectively 20% for the square) were inaccurate (r < 0.3, p < 0.05).
Nevertheless, mobile phone measurements are in agreement for the LA50 and the LA90 acoustic indicators
with the fixed station (4-m high) measurements, with a median deviation smaller than 1.5 dB(A) for the
boulevard (respectively 3 dB(A) for the square).
Ó 2016 Elsevier Ltd. All rights reserved.

1. Introduction
Urban sound environment is an important element for the def-
inition of the quality of life. Since 2002 and the European Directive
2002/49/EC, cities with >100,000 inhabitants must publish noise
maps on a regular basis [1]. These maps show the LDEN indicator
and are used for policy support and to communicate with citizens
about noise. Noise maps are currently generated out of a sparse
measurement basis using complex processing steps including sim-
ulations. Increasing the number of measurement points is a good
way to improve the quality and accuracy of noise maps [2,3]. How-
ever, a noise measurement network is expensive and complicated
to upgrade.
On the other hand, the widespread use of small and powerful
mobile devices enables a person-centric collection of environmen-
⇑ Corresponding author.
E-mail address: catherine.lavandier@u-cergy.fr (C. Lavandier).
tal data giving rise to the idea of participatory sensing [4,5]. Thanks
to the integrated sensors (such as microphone and GPS) and the
availability of broadband internet access smartphones can be used
as low cost sound level monitoring tools. In 2010, smartphones
already surpassed personal computer sales evidencing that mobile
devices were becoming fully integrated into everyday activities
supplanting existing computers as internet devices. Half of the
world’s population now has a mobile subscription and the increase
is very fast: from 3.6 billion in 2014 to 4.6 billion expected unique
subscribers in 2020 [6].
In this context, one particularly interesting application in envi-
ronmental acoustics is to collect participatory sound measure-
ments. Thus, several government and research organizations
have already commissioned noise pollution monitoring studies
using mobile phones [7–10]. As a result, there has been consider-
able progress in technologies for participatory mobile sound mea-
surements, giving rise to the development of several applications
such as SoundCity [11], EarPhone [12], NoiseSPY [13], and Noise-
Tube [14]. Many research efforts have then focused on assessing

http://dx.doi.org/10.1016/j.apacoust.2016.07.011
0003-682X/Ó 2016 Elsevier Ltd. All rights reserved.
220 P. Aumond et al. / Applied Acoustics 117 (2017) 219–226
the accuracy of the data gathered with these technologies inside
and outside a laboratory [15–19]. However, it is also crucial to
study and compare the results obtained during an entire large-
scale participatory sensing field campaign with reference measur-
ing methods such as noise-monitoring station measurements.
This paper studies the relevance and accuracy of mobile phones
for measuring urban noise pollution in a large-scale participatory
sensing campaign. In particular, the noise mapping campaign,
organized in the context of the Cart-ASUR project (French acronym
for: Cartographic representation of the quality of the urban sound
environments) in Paris for a year is analyzed. The goal was to
develop a cartographic indicator of urban sound quality, in terms
of a variety of perceptive data together with the evolution of the
sound level [20,21]. In order to evaluate the relevance of the
mobile sound measurements, noise measurements were taken
using both mobile phone technology and nearby sound monitoring
stations. Instantaneous A-weighting sound level, energy indicators
such as LA,eq, LA10, LA50 or LA90 and event indicators such as the
number of noise events exceeding a certain threshold La
(NNEL P La) could then be compared with a reference measure-
ment method of urban environmental noise.
The structure of this paper is as follows: First, an overview of
the technical tools is presented (mobile phones, and application)
and the procedure for correction of systematic errors by way of
laboratory calibration. In Section 3, two 10-min smartphone mea-
surement sessions are compared with a standardized sound level
meter and with a 4-m high monitoring station at two locations.
Then, an all-year measurements comparison with the noise-
monitoring station is presented (n = 59). To conclude, the validity
and accuracy of using mobile phones in a participatory noise mea-
surement campaign is discussed in Section 4, where a general per-
spective on potential improvements is proposed.
2. Experimental setup
The technical setup and calibration methodology is described in
this section. After this phase of calibration and application devel-
opment, a sound measurement campaign was performed in Paris
between September 2013 and September 2014. During one year,
a total of 3409 assessments were carried out with 60 mobile
phones at 28 selected locations in Paris during up to five homoge-
nous periods (day, evening, night, summer, winter). Each situation
(one location at one homogenous period) was assessed by about 20
people. A specific mobile phone model was given to the 60 partic-
ipants and internet and communication subscription was offered
for a year in order to control the experimental procedure to a
required extent. Before each perceptive assessment, sound pres-
sure levels were recorded each second during a 10-min period.
2.1. Noise measurement software
A specific mobile phone application was developed, based on
the NoiseTube open-source application [14] (see Fig. 1). With the
NoiseTube application, it is possible to record the sound pressure
level every second and to calculate the equivalent sound level in
dB(A). In addition to noise levels, the new Cart_ASUR application
collects perceptive data in specific places (parks, squares, thor-
oughfares, streets, schools, markets, pedestrian streets, etc.). The
application has been developed to respect to a great extent the
European measurements standards (A-filtering, direct read-out of
sound levels in dB(A), computation of LA,eq over arbitrary time
intervals, calibration, spherical sensitivity and wind protection)
[22]. The sound pressure level is recorded for each second from
the mobile phone’s microphone during a 10-min period. It has
been stated in earlier research [23] that this duration is long
enough to characterize the acoustic environment of an urban situ-
ation. From these measurements 20 acoustic indicators, often used
in environmental acoustics, are calculated as in [24]. In this study,
the accuracy of the following metrics is assessed: LA1s, LAeq,10min, LAx
for x = 5, 10, 50, 90, 95, r and LA10–LA90 as well as the event indica-
tors, Number of Noise Event, NNEL>y and NNEL>LAeq,10min+z, and Mask
Index, MIL>y and MIL>LAeq,10min+z with y = 70, 75 and 80 dB(A) and
z = 10 and 15 dB(A). Consequently all the measurements were
stored in real-time on servers, accessible via internet. Also, mea-
surement quality and experiment progress could be checked dur-
ing the entire campaign. If repetitive occurrence of erroneous
results were observed, the concerned user was contacted for
verification.
2.2. Smartphone calibration methodology
The mobile phone calibration methodology was realized by
BrusSense Team of Vrije Universiteit Brussel (VUB) respecting the
principles stated in [16,25]:
(1) The calibration was only carried out on the global dB(A), tak-
ing a pragmatic stance that sound levels have the most
important impact in urban environment and this unit is
the one used in most of noise environmental regulations.
However it was also considered that in the frequency and
the level ranges where urban sounds are dominant, the
mobile phone behaves generally linearly (as shown in
Fig. 2 for the mobile phone chosen in this study).
(2) As different mobile phone models show different behavior,
even in the same brand, the calibration parameters must
be calculated for each model.
(3) The calibration realized on one mobile phone model is
approximately suitable for all the phones of the same model,
even if the history of utilization from the phones differs.
White noise was used as a reference source to perform the cal-
ibration. This was generated in a anechoic chamber from 35 to
100 dB(A) by 5 dB(A) steps. All the phones were tested at the same
time, with the microphone facing towards the loudspeaker (at 1 m
distance) as well as the referenced microphone. The directivity
influence has not been evaluated. Each step was maintained for
at least 30 s in order to stabilize the noise level measurement. Cal-
ibration was performed by a comparative study between the
assessed mobile phones and a reference class 1 sound level meter
model CEL-500.
2.3. Mobile phone model choice
At the outset, a bench test was conducted in order to choose the
mobile phone model to use during the campaign. A relevant factor
for choosing the model was the fact that the mobile phones were
going to be used for measuring noise in an urban environment.
Two mobile phone models were sent to the laboratory in response
to a public tender with demanded specifications, such as price and
also microphone sensitivity, internal memory, cut off frequencies,
battery autonomy and Android operating system. The measure-
ments were done with the Noise Tube Application. Results showed
that the HTC One X was best suited for this study as it showed a
good sensitivity in the 35–100 dB(A) range. The tested Samsung
Galaxy S3 was more sensitive at low sound levels but showed a
low sensitivity around 90 dB(A), which could be problematic in
an urban environment context (see Table 1).
After having selected the HTC One X model, its frequency
response was then measured and compared to a reference G.R.A.
S. MCE 212 class 1 microphone in a small semi-anechoic chamber
and with a Genelec 1031A loudspeaker. Fig. 2 shows that the
 P. Aumond et al. / Applied Acoustics 117 (2017) 219–226 221

Fig. 1. Screenshots of the Cart_ASUR application. (a) Recording of the sound pressure level, (b) part of perceptive survey, (c) resume of sound measurements; more details
given in [20].

Fig. 2. Sensitivity response of mobile phone HTC One X using reference G.R.A.S. Class 1 microphone.

Table 1
Primary bench test between two mobile phone models from 35 to 100 dB(A). Relative level to the previous step in dB(A).

Ref. levels - dB(A) 35 40 45 50 55 60 65 70 75 80 85 90 95 100

Class1 sound level meter Ref +5 +5 +5 +5 +5 +5 +5 +5 +5 +5 +5 +5 +5
HTC One X Ref +0.3 +0.8 +1.6 +2.5 +4 +5 +4.9 +5.2 +4.4 +5.2 +5.4 +4.8 +5.1
Samsung Galaxy 3 Ref +3.8 +6.3 +6.9 +6.2 +5.4 +3.8 +4.9 +4.7 +5.5 +1.5 +1.3 +0.2 +0.3

mobile phone deviates from the referenced microphone above
5 kHz and below 100 Hz. Also, between 63 Hz and 10 kHz, the sen-
sitivity is independent of the level of measurements (between
40 dB and 95 dB) as generally observed for other company mobile
phones [16,25]. These results permit to justify working only with
global values as presented in Section 2.2.
2.4. Mobile phone intervariability and systematic error correction
In order to assess that the calibration data was approximately
suitable for all phones used in the campaign, seven mobile phones
of the HTC One X model (MP1 to MP7), were randomly chosen to
be calibrated. In order to check mobile phone stability over time,
two of them (MP6 and MP7), were tested one year later, just before
the beginning of the participative campaign.
In Fig. 3, one can observe that all mobile phones behaved sim-
ilarly and linearly in the range between 50 and 90 dB(A). Following
the average systematic error of the seven mobile phones, a calibra-
tion vector was determined. Four of the seven tested mobile
phones, showed very similar behavior below 55 dB(A) (MP1,
MP2, MP6 and MP7). The calibration vector below 55 dB(A) is
slightly adjusted in terms of the average values of these four
mobiles phones assuming that these phones represent the stan-
dard behavior of the 60 phones used. Nevertheless, in this zone,
as the variability between the phones was higher, it could be
expected that the accuracy would be lower.
Systematic errors for the HTC One X phone were corrected
using a calibration vector presented in Table 2. This correction
was applied at each second to the sound pressure level measure-
ment. If the measured value was located between two calibration
222 P. Aumond et al. / Applied Acoustics 117 (2017) 219–226

Fig. 3. Comparison between measured sound levels in dB(A) for seven not calibrated mobile phones (MP1 to MP7) and one calibrated mobile phone (MP8) with a reference
Class1 sound level meter.

Table 2
Average original and adjusted sound levels measured with 5 mobiles phones over a 30.1–93 dB(A) range.

Levels dB(A) 30.1 35.1 40.1 45 50 55 60.1 65 70 75.1 80.3 85 90.1

Av. original level 27.5 27.8 28.8 30.4 33.4 37.0 41.7 46.3 51.3 56.3 61.7 66.6 74.5
Calibration vector +2.6 +7.3 +11.3 +14.6 +16.6 +18 +18.4 +18.7 +18.7 +18.8 +18.6 +18.4 +15.6
Av. corrected level 30.1 35.1 40.1 45 50 55 60.1 65 70 75.1 80.3 85 90.1

points, a linear regression between these points was operated to
determine the new corrected level.
In order to check the relevance of the calibration for a new
mobile by cross-validation, the last test was done with another
randomly chosen mobile phone (MP8) calibrated with the system-
atic error corrections, (see Fig. 3). For this phone, the adjusted
sound level measurement accuracy, in laboratory condition, was
less than 5 dB(A) and below 1 dB(A) in the 50–90 dB(A) range.
Another way to observe the quality of the mobile phone mea-
surements is to look at their variability. In Fig. 4, the interquartile
range between all tested phones (MP1-MP8) is shown after appli-
cation of the calibration vector. Below 50 dB(A), the measured
value can be associated with an interquartile range of 8 dB(A),
approximatively ±4 dB(A), while above this limit the interquartile
range is around 3 dB(A), approximatively ±1.5 dB(A).
Note that less accuracy below 50 dB(A) and above 90 dB(A) is
not a major issue for this campaign since measurements take place
in urban areas, where sound levels are mostly within this range
[26]. These conclusions are similar to the ones presented for the
same calibration process with a Nokia 5230 phone model in prior
work [16].
2.5. Operator error
In order to minimize the operator error source, focus groups
with all the participants were organized. During these focus group
meetings, information was given about the location of the micro-
phone on the phone. The participants were told how to hold the
phone in front of them, with a minimum 20 cm distance from their
body. Silence and a steady position should be maintained. More-
over, when a series of erratic measurements was detected (for
example a wrong location or a wrong moment, a lot of seconds
during a measurement when sound levels are around 0 dB(A), or
a measurement where the time evolution of the sound level is
under 10 dB(A) with punctual high values), the involved partici-
pant was contacted by e-mail in order to make him/her aware of
the problem, and the corresponding data was discarded. The
review of the measurements taken by the participants was carried
out manually every week. Apart from detecting the erratic mea-
surements, the regularity of participation was also verified. During
this one year campaign, more than 300 e-mails were sent to the
participants and to the operator in order to resolve problems
(insurance, robbery, breakage, etc.). With this intricate monitoring,
the data selected for analysis can be considered as relevant.

3. Evaluation of smartphone sound measurements

3.1. Inter-comparison between mobile phone, sound level meter and

fixed stations

Fig. 4. Inter-variability between the tested mobile phones after application of the This work aims to answer whether acoustic measurements
calibration vector (Q3-Q1 of the assessed sound level by phones). delivered by calibrated mobile phones can be similar and corre-
 P. Aumond et al. / Applied Acoustics 117 (2017) 219–226 223

lated with monitoring station measurements, commonly used for
overall documentation of environmental noise. To verify this
assumption, two LadybirdÓ (Azimut Monitoring) fixed monitoring
stations were installed for the duration of the experiment on street
lamps at 4 m height at two different locations: the boulevard Ras-
pail (6 months) and the square Gilbert Perroy (6 months) were
chosen as reference comparison points (see Fig. 5).
In order to observe measurement point height influence, a class
1 sound level meter (model RION NL52) measured sound levels
close to a tested mobile phone. Fig. 6 shows the 10-min sound
measurement time series in the boulevard and square locations
with the sound level meter (in yellow), with a fixed station (in
red) and with one randomly-chosen HTC One X mobile phone (in
blue).
From the results in Table 3, a strong correlation is observed
between the three measurement systems for both situations. The
correlation coefficients are stronger for the boulevard than for
the square. Since the correlation is stronger between the phone
and the sound level meter, than between the fixed station and both
other systems, it can be assumed that this difference is explained

(a) (b)
Fig. 5. Photography onsite. (a) boulevard location (Boulevard Raspail, Paris), (b) square location (Square Gilbert Perroy, Paris).

Fig. 6. Sound level measurements time series in boulevard (a) and square (b) locations with sound level meter, fixed station and mobile phone.

Table 3
Comparison of correlation coefficient (⁄⁄p < 0.05, ⁄p < 0.1), RMSE between fixed stations, sound level meter and phone for 600 measures of LAeq,1s.

Boulevard Square Both

Fixed Station Mobile Phone Fixed Station Mobile Phone Fixed Station Mobile Phone
Sound level meter r = 0.99⁄⁄ r = 0.98⁄⁄ r = 0.80⁄⁄ r = 0.96⁄⁄ r = 0.98⁄⁄ r = 0.97⁄⁄
RMSE = 1.49 RMSE = 3.6 RMSE = 2.5 RMSE = 3.0 RMSE = 2.0 RMSE = 3.4
Fixed Station r = 0.98⁄⁄ r = 0.80⁄⁄ r = 0.96⁄⁄
RMSE = 3.2 RMSE = 2.3 RMSE = 2.8
224 P. Aumond et al. / Applied Acoustics 117 (2017) 219–226

by the larger distance to the sound sources for the fixed measure-
ment station, especially for human sources in the square.
Table 4 presents the calculated indicator values, integrated over
10 min for both situations and for all measurement systems. An
overestimation of the mobile phone sound levels limited to 3 dB
(A) above 70 dB(A) and an underestimation limited to 2 dB(A)
below 60 dB(A) compared to both fixed station and class 1 sound
level meter are observed. Thus, mobile phones show an acceptable
accuracy compared to both other systems. Other indicators, such as
standard deviation or event numbers and their duration are also
assessed in an acceptable way. It is worth noticing that the number
of event (NNE) is overestimated by the mobile when the location
equivalent sound level is close to the selected threshold (70 dB
(A) threshold for the square, respectively 80 dB(A) for the boule-
vard). This is due to a slight overestimation of the mobile com-
pared to the others equipment’s at these levels. Finally, it is
observed that the mobile phone measurements are able to high-
light square and boulevard sound level distribution differences.
3.2. Full measurement campaign comparison
A one-year campaign was carried out with 60 participants. Half
of them were instructed to carry out measurements below one of
the two dedicated fixed monitoring stations. In total, 59 measure-
ments were taken, 30 on the boulevard, and 29 in the square. As
shown in Fig. 7, 10% of the measurement had a very erratic behav-
ior (r < 0.35). This can be explained by a phone failure or if mishan-
dled or misused by the user. When these kind of measurements
was detected, the participants were instructed to be more careful.
Further, 50% of the mobile phone square measurements presented
a medium correlation (r > 0.5) with the monitoring station and 50%
of the boulevard mobile phone measurements presented a strong
correlation (r > 0.75). Errors were probably due to three principal
reasons: (1) The measurement height difference implied that
mobile phones were more sensitive to sound sources such as
voices, which are particularly present in a square. (2) The general
sound level was lower in the square than in the boulevard (see

Table 4
Measured 10-min indicators by phone, sound level meter and fixed monitoring station on Boulevard and Square locations.

Boulevard Square
Fixed station Sound level meter Mobile Phone Fixed station Sound level meter Mobile Phone
LAeq dB(A) 72.2 72 75.4 65.1 63.4 66.5
LA5 dB(A) 77.4 77.5 81 68.4 67 70
LA10 dB(A) 76 75.5 79 67.3 66.1 69
LA50 dB(A) 69.8 68.9 73 64.2 62.4 65
LA90 dB(A) 53.4 51.9 52 60.8 58.9 62
LA95 dB(A) 51.4 50.1 49 59.9 57.6 60
LA10-LA90 dB(A) 22.6 23.6 27 6.5 7.3 7
r dB(A) 8.4 8.8 10 2.7 2.9 3.1
NNEL>70 dB(A) 28 37 19 6 3 22
NNEL>75 dB(A) 30 38 44 2 0 2
NNEL>80 dB(A) 5 7 24 0 0 0
NNEL>L10 dB(A) 24 31 34 32 28 27
NNEL>LAeq,T+10 dB(A) 2 3 2 2 0 0
NNEL>LAeq,T+15 dB(A) 0 1 0 0 0 0
MIL>70 dB(A) s 300 284 360 15 6 66
MIL>75 dB(A) s 125 119 261 4 0 4
MIL>80 dB(A) s 12 16 76 0 0 0
MIL>L10 dB(A) s 85 91 112 96 90 116
MIL>LAeq,T+10 dB(A) s 4 6 4 4 0 0
MIL>LAeq,T+15 dB(A) s 0 2 0 0 0 0

Fig. 7. Cumulative distribution of the correlation coefficient between 600 LAeq,1s measured by phone and monitoring station on boulevard (30 repetitions) and square location
(29 repetitions). All are significant at 95% level, except for r < 0.1.
 P. Aumond et al. / Applied Acoustics 117 (2017) 219–226 225

Fig. 8), which induced higher sensitivity to user generated noises.
(3) The sound level variation was larger in the boulevard than in
the square (see Fig. 8), which induced a higher correlation
coefficient.
Fig. 8 presents the relationship between LAeq,1s as measured by
phones and the monitoring station for boulevard and square loca-
tions during the whole experiment. The phone measurements tend
to overestimate the sound level in the square and for the highest
sound level.
In order to evaluate the use of mobile phones in an urban con-
text, correlation coefficients of the calculated 10-min indicators
between phone and monitoring stations, are presented in Table 5.
The indicators r, LA10–LA90, LA50, LA90, LA95 are significantly cor-
related (p < 0.05) for both locations. Note that since the variation of
all the variables is small for one location, a weak correlation does
not mean that the accuracy is necessarily weak. One also observes
most of the indicators are assessed with a median error of 1.5 dB
to 3 dB for LA95, LA90 and LA50. Dynamic indicators like r or LA10–
LA90 are assessed appropriately. It can be noticed that LA50 is
assessed better than LAeq, 10min as it is less sensitive to sound level
peaks.
The energetic indicators measured by mobile phone in boule-
vard and square locations (Table 4) are able to reveal the difference
between both sound environments, and give a relevant assessment
of these indicators compared to the fixed stations.
4. Discussion
This study was performed under controlled conditions. Only
one specific company and mobile phone model has been chosen
and used for the overall measurement campaign. A specific calibra-
tion limited measurement errors of the mobile phones. This cali-
bration was carried out using a bench test with seven of the
whole fleet (60 mobiles) by correcting systematic errors in sound

Fig. 8. Comparison between phone and fixed monitoring station measurement on square (a) and boulevard (b) locations. The linear gray scale represents the normalized
number of measurements.

Table 5
Median, 1st and 3rd quartile range of the difference (D) for mobile and fixed station measured 10-min indicators.

10-min indicators Median, 1st and 3rd quartiles of D (Mobile-Fixed Correlation r (Mobile-Fixed Station) (⁄⁄p < 0.05,
⁄
Station) p < 0.1)
Boulevard (30 pts) Square (29 pts) Boulevard (30 pts) Square (29 pts)
LAeq dB(A) 2.7 (2.6; 3.6) 4.2 (2; 6.4) 0.37⁄ 0.19
LA5 dB(A) 3.0 (0.7; 4.5) 4.3 (2.5; 7.2) 0.20 0.06
LA10 dB(A) 2.5 (0.4; 3.4) 3.4 (1.6; 6.2) 0.18 0.22
LA50 dB(A) 1.5 ( 0.1; 2.1) 2.8 (1.7; 5.2) 0.70⁄⁄ 0.41⁄⁄
LA90 dB(A) 0.4 ( 1.3; 2.3) 2 (0.8; 4.5) 0.56⁄⁄ 0.66⁄⁄
LA95 dB(A) 0.1 ( 1.3; 2.4) 1.8 (0.1; 4.5) 0.52⁄⁄ 0.69⁄⁄
LA10-LA90 dB(A) 1.4 ( 0.8; 2.8) 1.7 (0.2; 2.7) 0.64⁄⁄ 0.77⁄⁄
r dB(A) 0.8 (0.2; 1.1) 0.9 (0.42; 1.36) 0.58⁄⁄ 0.68⁄⁄
NNEL>70 dB(A) 15 (9; 19) 15 (9; 20) 0.33⁄ 0.42⁄⁄
NNEL>75 dB(A) 4 (2; 14) 4 (2; 14) 0.07 0.09
NNEL>80 dB(A) 1 (0–3) 1 (0–4) 0.20 0.04
NNEL>L10 dB(A) 6.5 ( 1-10) 7 (0 11) 0.21 0.34⁄
NNEL>LAeq.T+10 dB(A) 1.5 (0–3) 1.5 (0–3) 0.08 0.25
NNEL>LAeq.T+15 dB(A) 0 (0 0) 0 (0 0) 0.07 0.16
MIL>70 dB(A) s 76 (51–212) 76 (41–216) 0.61⁄⁄ 0.13
MIL>75 dB(A) s 16 (5–46) 16 (4–50) 0.32 0.13
MIL>80 dB(A) s 2 (0–16) 2 (0–17) 0.22 0.09
MIL>L10 dB(A) s 19 (7–34) 19 (9–34) 0.16 0.28
MIL>LAeq.T+10 dB(A) s 4 (0 12) 4 (0 12) 0.19 0.22
MIL>LAeq.T+15 dB(A) s 0 (0 0) 0 (0 0) 0.07 0.13
226 P. Aumond et al. / Applied Acoustics 117 (2017) 219–226
level measurements. Moreover, the frequency response of the
selected mobile was tested in comparison to a class 1 reference
microphone. It revealed that the mobile phone deviates from the
referenced microphone above 5 kHz and below 100 Hz. Most
smartphones integrate MEMS microphone, including HTC One X.
These microphones often have flatter frequency response and
wider noise sensitivity than the observed in this experiment. This
is due to the fact that the Cart-ASUR application did not bypassed
the built-in filters which are typically integrated in smartphones.
Those filters alter the recorded signals and are optimized for voice
transmission [15]. In future work, we will remove these filters.
In addition, focus groups were organized, with all participants,
and information was given about microphone position and correct
measurement behavior. Participants were contacted if repeated
erratic measurements were detected, and the corresponding data
was then discarded. This requires a strict organization which can-
not be respected for a large public participatory experiment. Nev-
ertheless, using different mobile phone models with different
calibration patterns could avoid repeating the same systematic
errors. In a participatory experiment, the use of a large number
of phones would allow the use of statistical methods for eliminat-
ing erratic measurements as proposed in [27]. Finally, if measure-
ment positions are not controlled, it is necessary to develop filters
to detect phone misusage (hand or pocket location, operator speak-
ing, etc.) [8,28].
5. Conclusion
In this paper the accuracy and relevance of mobile technology
for a large-scale participatory sensing campaign has been studied.
The Cart_ASUR participatory noise pollution monitoring campaign
was conducted in Paris for one year using 60 mobile phone devices.
The main results can be summarized as follows:
(1) After a calibration phase, mobile measurements for a 10-min
duration are similar to those realized by a class 1 micro-
phone or a fixed station for most of the typical energetic
and event indicators used in environmental acoustics. The
median difference of the overall LA50 or LA90 measured by
mobile phones or by fixed stations installed at 4-m height
is limited to 1.5 dB(A) in a boulevard sound environment,
and to 3 dB(A) in a square. Nevertheless, 10% of the mea-
surements along the boulevard (respectively 20% in the
square) showed worse accuracy (r < 0.35) due to phone fail-
ure or misusage.
(2) Instantaneous sound levels with mobile phones can show a
very strong correlation (r > 0.9) with reference class 1 sound
level meter measurements, with a root mean square error
(RMSE) smaller than 3 dB(A).
This study showed that, with intermittent monitoring and a
careful review of the collected data, the mobile phone can be a very
useful tool as it permits to combine a large number of perceptive
assessments with acoustic measurements.
Acknowledgment
This work has been funded by the French Environment and
Energy Management Agency (ADEME) under the convention n
°1217C0035.
References
[1] EC. Directive 2002/49/EC of the European parliament and the council of 25
June 2002 relating to the assessment and management of environmental noise.
Off J Euro Commun 2002;189(12):12–25.
[2] Zannin PHT, Engel MS, Fiedler PEK, Bunn F. Characterization of environmental
noise based on noise measurements, noise mapping and interviews: a case
study at a university campus in Brazil. Cities 2013;31:317–27.
[3] Mioduszewski P, Ejsmont JA, Grabowski J, Karpiński D. Noise map validation
by continuous noise monitoring. Appl Acoust 2011;72(8):582–9.
[4] Burke JA, Estrin D, Hansen M, Parker A, Ramanathan N, Reddy S, et al.
Participatory sensing. In: Proceedings of the international workshop on world-
sensor-web (WSW’2006), co-located with ACM SenSys. Boulder, USA; 2006.
[5] Kim S, Mankoff J, Paulos E. Sensr: evaluating a flexible framework for authoring
mobile data-collection tools for citizen science. In: Proceedings of the 2013
conference on computer supported cooperative work - CSCW’13. p. 1453.
[6] GSMA Mobile Economy. The mobile economy - executive summary; 2015.
[7] Maisonneuve N, Stevens M, Ochab B. Participatory noise pollution monitoring
using mobile phones. Inf Polity 2010;15:51–71.
[8] Rana R, Chou CT, Bulusu N, Kanhere S, Hu W. Ear-phone: a context-aware noise
mapping using smart phones. Pervas Mob Comput 2014;17:1–22.
[9] Bocher E, Picaut J, Guillaume G, Petit G, Fortin N, Can A, et al. Crowdsourcing of
noise map pollution using smartphones; June 2015.
[10] Aspuru I, García I, Herranz K, Santander A. CITI-SENSE: methods and tools for
empowering citizens to observe acoustic comfort in outdoor public spaces.
Noise Mapp 2016;3(1):37–48.
[11] Inria. Sound cityAvailable from: <http://www.inria.fr/centre/paris-
rocquencourt/actualites/lancement-de-l-appli-soundcity>2015 [accessed: 02-
December-2015].
[12] Rana RK, Chou CT, Kanhere SS, Bulusu N, Hu W. Ear-phone: an end-to-end
participatory urban noise mapping system. In: Proceedings of the
international conference on information processing in sensor networks IPSN.
p. 105–16.
[13] Kanjo E. NoiseSPY: a real-time mobile phone platform for urban noise
monitoring and mapping. Mob Networks Appl 2010;15:562–74.
[14] Maisonneuve N, Stevens M, Niessen ME, Steels L. NoiseTube: measuring and
mapping noise pollution with mobile phones. Environ Sci Eng (Subseries
Environ Sci 2009:215–28.
[15] Kardous CA, Shaw PB. Evaluation of smartphone sound measurement
applications. J Acoust Soc Am 2014;135(4):EL186–92.
[16] D’Hondt E, Stevens M, Jacobs A. Participatory noise mapping works! An
evaluation of participatory sensing as an alternative to standard techniques for
environmental monitoring. Pervasive Mob Comput 2013;9(5):681–94.
[17] Von Kaenel M, Sommer P, Wattenhofer R. I.R. Large-scale participatory sensing
at high altitudes. In: Proceedings of the 12th workshop on mobile computing
systems and applications - HotMobile’11; 2011. p. 63.
[18] Neitzel RL, Heikkinen MSA, Williams CC, Viet SM, Dellarco M. Pilot study of
methods and equipment for in-home noise level measurements. Appl Acoust
2016;102:1–11.
[19] Ibekwe TS, David FO, Dahilo Ea, Gbujie IO, Nwegbu MM, Nwaorgu OG.
Evaluation of mobile smartphones app as a screening tool for environmental
noise monitoring. J Occup Environ Hyg 2015;9624(September):1–20.
[20] Ricciardi P, Delaitre P, Lavandier C, Torchia F, Aumond P. Sound quality
indicators for urban places in Paris cross-validated by Milan data. J Acoust Soc
Am 2015;138(4):2337–48.
[21] Lavandier C, Delaitre P, D’Hondt E, Boix E Gonzalez, Kambona K. Urban sound
quality assessment with mobile technology: the Cart_ASUR project. In:
Acoustics 2013; 2013.
[22] European Commission. Commission recommendation of 6 August 2003
concerning the guidelines on the revised interim computation methods for
industrial noise, aircraft noise, road traffic noise and railway noise, and related
emission data. Off J Euro Union L 2003;212:49–64.
[23] Brocolini L, Lavandier C, Quoy M, Ribeiro C. Measurements of acoustic
environments for urban soundscapes: choice of homogeneous periods,
optimization of durations, and selection of indicators. J Acoust Soc Am
2013;134(1):813–21.
[24] Can A, Leclercq L, Lelong J, Defrance J. Capturing urban traffic noise dynamics
through relevant descriptors. Appl Acoust 2008;69(12):1270–80.
[25] Stevens M. Community memories for sustainable societies: the case of
environmental noise [Doctoral Dissertation]. Vrije Universiteit Brussel; 2012.
[26] Brocolini L. Caractérisation de l’environnement sonore urbain: proposition de
nouveaux indicateurs de qualité [Thèse de Doctorat]. Université de Cergy-
Pontoise; 2012.
[27] Can A, Guillaume G, Picaut J. Cross-calibration of participatory sensor
networks for environmental noise mapping. Appl Acoust 2016;110:99–109.
[28] Yang J, Munguia-Tapia E, Gibbs S. Efficient in-pocket detection with mobile
phones. In: Proceedings of the 2013 ACM conference on pervasive and
ubiquitous computing adjunct publication - UbiComp’13 adjunct. p. 31–4.