Similarity and pleasantness assessments of water-fountain

sounds recorded in urban public spaces

Maria Radsten Ekman
G
osta Ekman Laboratory, Department of Psychology, Stockholm University, SE-10691 Stockholm, Sweden

n
Peter Lunde
Department of Sustainable Built Environment, SP Technical Research Institute of Sweden, Box 857,
SE-50115 Boras, Sweden

Mats E. Nilssona)
G
osta Ekman Laboratory, Department of Psychology, Stockholm University, SE-10691 Stockholm, Sweden

(Received 12 March 2015; revised 14 September 2015; accepted 18 October 2015; published online
16 November 2015)
Water fountains are potential tools for soundscape improvement, but little is known about their
perceptual properties. To explore this, sounds were recorded from 32 fountains installed in urban
parks. The sounds were recorded with a sound-field microphone and were reproduced using an
ambisonic loudspeaker setup. Fifty-seven listeners assessed the sounds with regard to similarity and
pleasantness. Multidimensional scaling of similarity data revealed distinct groups of soft variable
and loud steady-state sounds. Acoustically, the soft variable sounds were characterized by low
overall levels and high temporal variability, whereas the opposite pattern characterized the loud
steady-state sounds. The perceived pleasantness of the sounds was negatively related to their
overall level and positively related to their temporal variability, whereas spectral centroid was
weakly correlated to pleasantness. However, the results of an additional experiment, using the same
sounds set equal in overall level, found a negative relationship between pleasantness and spectral
centroid, suggesting that spectral factors may influence pleasantness scores in experiments where
overall level does not dominate pleasantness assessments. The equal-level experiment also showed
that several loud steady-state sounds remained unpleasant, suggesting an inherently unpleasant
sound character. From a soundscape design perspective, it may be advisable to avoid fountains
C 2015 Author(s). All article content, except where otherwise noted, is
generating such sounds. V
licensed under a Creative Commons Attribution 3.0 Unported License.
[http://dx.doi.org/10.1121/1.4934956]
[KVH]

Pages: 30433052

I. INTRODUCTION

Water fountains are installed in many urban public

spaces. In addition to their visual qualities, fountains may
generate pleasant sounds, thereby improving the quality of
the acoustic environment or soundscape. However, perceptual studies of water-generated sounds suggest great variation in listener preferences. For example, sounds from
natural streams and sea waves tend to be perceived as pleasant, whereas sounds from waterfalls tend to be perceived as
unpleasant (e.g., Radsten-Ekman et al., 2013; Galbrun and
Calarco, 2014). This suggests that the auditory aspects of
water fountains should be considered, as an unpleasant
sound may counteract the positive visual effects a fountain
have on the quality of its location. This study evaluated perceived similarity and pleasantness of a large set of urban
water fountains. The purpose was to identify perceptual
dimensions of sounds generated by the fountains and to
assess correlations between perceived pleasantness of the
sounds and their basic acoustic properties.

a)

Electronic mail: mats.nilsson@psychology.su.se

J. Acoust. Soc. Am. 138 (5), November 2015

Water features add to the visual attractiveness of areas

(Nasar and Li, 2004; Dramstad et al., 2006; White et al.,
2010). Listening experiments verify that adding water-fountain
sounds may increase the overall quality of the soundscape
(Jeon et al., 2010, 2012), but care must be taken as unpleasant
water sounds may detract from the overall soundscape quality
(Radsten-Ekman et al., 2013). The ability of water-generated
sounds to partially or completely mask unwanted sounds, such
as road traffic noise, is limited to situations in which the two
sources have similar temporal and spectral characteristics
(Watts et al., 2009; Nilsson et al., 2010; De Coensel et al.,
2011; Galbrun and Ali, 2013). Road traffic noise typically contains sizeable low-frequency components; this makes many
water-generated sounds ill-suited as maskers, as their spectra
are dominated by higher-frequency energy. However, some
water features with high flow rates may generate broadband
sounds that can partially or completely mask road traffic noise
(Galbrun and Ali, 2013). A problem, however, is that such
water features may themselves generate unpleasant sounds; for
example, studies have suggested that waterfall-like structures
generate sounds considerably less pleasant than those of water
features with lower flow rates (Radsten-Ekman et al., 2013;
Galbrun and Calarco, 2014).

0001-4966/2015/138(5)/3043/10

C Author(s) 2015
V

3043

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 188.24.122.154 On: Tue, 08 Mar 2016 15:45:56

Fountains have rising jets, downward falls, or a combination of the two. Factors influencing their sound include
flow rate, falling water height, impact materials, and number
of jets (Watts et al., 2009; Galbrun and Ali, 2013). These
factors are related to various acoustic properties of the
sounds. Previous studies have explored various acoustic indicators that may predict the perceptual properties of fountain
sounds. Overall level, the main determinant of perceived
loudness, is a main factor, as high loudness is associated
with low preference of water-generated sounds (RadstenEkman et al., 2013), and, generally, with high annoyance of
environmental noise (e.g., Berglund et al., 1990). Watergenerated sounds with high temporal variability have been
found to be more pleasant than sounds with a steady-state
character (Galbrun and Ali, 2013). The role of spectral envelope is less clear. For example, both negative (Galbrun and
Ali, 2013) and positive relationships (Watts et al., 2009;
Jeon et al., 2012) have been reported between preferences of
water-generated sounds and the psychoacoustic measure
sharpness, which is related to amount of high frequency content of sounds.
The purpose of this study was to evaluate a large set of
urban fountains recorded in public open spaces. Listeners
assessed the sounds in terms of their perceived similarity, to
obtain a representation of the sounds in perceptual space.
Similarity assessments are based on the perceived similarity
of sounds on salient or dominating perceptual dimensions
(e.g., Gygi et al., 2007). Thus, analyses of similarity data,
typically using multidimensional scaling (MDS), may give

insight into perceptually significant aspects of sounds, in the

present application a set of fountain-generated sounds. The
listeners also assessed the sounds in terms of their position
on the bipolar unpleasantpleasant dimension. This is the
main dimension of Axelssons circumplex model of soundscape quality (Axelsson et al., 2010), which is similar to previously proposed circumplex models of emotions (Russel,
1980), environments (Russel and Mehrabian, 1978), and
sound quality (Vastfjall et al., 2002), all of which propose a
fundamental likedislike or valence dimension (cf.
Kuppens et al., 2013). In the present study, unpleasantpleasant scores of sounds were related to acoustic measures of overall level, variability over time and spectral
envelope. In an additional experiment, all sounds were
adjusted to an equal overall sound pressure level (SPL) to
specifically explore the role of spectral and temporal properties for perceived pleasantness of water-fountain sounds.
II. METHOD
A. Fountain sound recordings

Twenty-eight recording sites were selected from an initial set of 61 sites in the Stockholm area. The main reasons
for excluding sites were that the fountains were turned off or
the presence of disturbing noise from construction work or
other noise sources. Eight of the chosen sites had more than
one fountain. In total, this resulted in recordings of 42 fountains. From these, 32 fountains, from 28 different sites, were
selected for the experiment. The recordings were selected to

FIG. 1. (Color online) Photos of a subset of the recorded fountains. Numbers

correspond to the fountain numbers in
the following result figures (rankordered from most pleasant, 1, to least
pleasant, 32).

3044

J. Acoust. Soc. Am. 138 (5), November 2015

Radsten Ekman et al.

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 188.24.122.154 On: Tue, 08 Mar 2016 15:45:56

obtain a large variety in fountain sounds and to exclude

recordings with prominent background sounds from people
talking, ventilation systems, road traffic, etc.
Each fountain sound was recorded with a four-channel
ambisonic microphone (Soundfield SPS200 microphone,
Marlow, UK) comprising four directional microphones in
a tetrahedral configuration and a single measurement
microphone (Br
uel & Kjr 4231 with conditioning amplifier type 5935, Nrum Denmark). All microphone outputs
were fed into a Sound Device 788T digital audio recorder
(Reedsburg, WI).
Recordings were made at the location nearest the fountain where visitors could be expected to stay or sit. The distance between microphone and fountain side was 0.51.5 m.
Figure 1 shows photos of several selected fountains.
Recording equipment is visible in the photo of fountain number 1. (The fountains are rank-ordered from the one generating the most pleasant sound, #1, to the one generating the
least pleasant sound, #32, based on the results presented
below, see Sec. III A.)
B. Acoustic analyses of experimental sounds

From each of the 32 fountain recordings, a 30-s excerpt

from the ambisonic recording was selected for the listening
experiment. The corresponding 30-s excerpts were extracted
from the one-channel measurement recordings and used for
the acoustic analyses. One of the fountain sounds (#20)
included a period of a slowly increasing level for about 7 s,
followed by a period of steady state level. Analyses of this
sound included only the steady-state period, as it could be
assumed that this dominated the listener responses to the
sound. Data analyses were also conducted with this sound
excluded and, unless otherwise stated, these analyses gave
results very similar to those presented below.
The experimental sounds were subjected to a variety of
analyses using the ArtemiS software, version 12 (HEAD
acoustics, Herzogenrath, Germany). In this article, the focus is
on results from analyses of A-weighted SPLs and narrow-band
spectra. Three basic measures were derived representing overall level, time variability, and spectral envelope. Overall level
was measured as the A-weighted equivalent continuous SPL
(LAeq,30s), variability was measured as the standard deviation
of instantaneous A-weighted SPLs (SDLA, time weighting
fast), and spectral envelope was measured as the spectral centroid (SC) of the 1/96-octave-band spectrum, that is, the frequency band for which the sum of lower band levels equals
the sum of higher band levels. In addition, Aures sharpness
(Aures, 1985) was calculated (unit: acum), as this psychoacoustic measure has been used in several previous studies
reporting both positive and negative relationships with preference of water-generated sounds (see Sec. I). Sharpness was
calculated from the sounds average spectra.
The analyzed sounds were high-pass filtered at 100 Hz
to reduce influence of ambient low-frequency components.
Analyses were also conducted using a 500 Hz high-pass filter, but these analyses yielded results very similar to those
presented below.
J. Acoust. Soc. Am. 138 (5), November 2015

Recordings, photos, and perceptual data are available

upon request to the third author.
C. Perceptual measures
1. Perceived similarity

A free sorting method was used to measure the perceived similarity of the fountain sounds. The instruction was
to sort sounds in groups based on perceived similarity using
as many groups as the listeners found appropriate (e.g.,
Coxon, 1999). The sounds were sorted using a software
application developed for this experiment. The listeners
could listen to a sound by clicking its icon, and then dragging the icon to any place on the screen. Groups were created by placing icons of similar sounds near each other on
the screen. The icons were assigned random numbers, which
differed between listeners. The listeners were free to listen to
each sound as many times as desired until a final sorting had
been achieved; they were then asked to verbally describe
what characterized the sounds in each sorted group of
sounds.
Each listener sorted the sounds once. The number of
times two sounds were sorted into the same group was used
as a measure of their perceived similarity. Two sounds were
duplicated, and the number of times a sound and its copy
was sorted into the same group was used as a measure of the
reliability of the sorting procedure. One of the duplicate
sounds was sorted into the same group by 50 listeners and
the other duplicate by 54 listeners, suggesting that most of
the 57 listeners reliably followed the sorting instructions.
Analyses of the data excluding the few listeners who did not
sort duplicates in the same group yielded very similar results
to those presented below, which were based on data from all
listeners.
2. Perceived pleasantness

After the sorting task, the listeners assessed the sounds

with regard to attributes defined by two orthogonal bipolar
dimensions, unpleasantpleasant versus uneventfuleventful,
that define Axelssons circumplex model of soundscape
quality (Axelsson et al., 2010). The focus of this article is on
the unpleasantpleasant dimension, since it is more relevant
for assessments of single sources than the uneventfuleventful dimension which is more relevant for assessments of
multi-source soundscapes. Listeners assessed the sounds on
the unpleasantpleasant dimension in two ways. First, using
a software application developed for this experiment, they
placed icons in an area of the screen defined by the orthogonal unpleasantpleasant (x-axis) and uneventfuleventful
(y-axis) scales. The icons were assigned random numbers
that differed between listeners. The listeners listened to a
sound by clicking its icon and were free to listen to each
sound as many times as desired. The final locations of the
sounds were saved as coordinates in the two-dimensional
space and the x-coordinate was used as the unpleasant
pleasant scale value. Second, in a separate session, the listeners assessed each fountain sound on four bipolar scales,
including an unpleasantpleasant scale with nine categories,
Radsten Ekman et al.

3045

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 188.24.122.154 On: Tue, 08 Mar 2016 15:45:56

scored from 4 through 0, to 4 (the other scales being

uneventfuleventful, chaotictranquil, and monotonous
exciting). Assessments on the bipolar unpleasantpleasant
scale agreed well with the unpleasantpleasant values derived
from the interactive method. Therefore, averages of scores
from the two methods were used to calculate the final unpleasantpleasant scores. To simplify the presentation, unpleasantpleasant scores will hereafter be called pleasantness
scores. Note that negative pleasantness scores refer to
unpleasant ratings on the bipolar scale.
D. Procedure

All listeners conducted the experiment in the same

order, starting with similarity sorting, followed by interactive pleasantness assessments, and, finally, pleasantness
assessments on a bipolar scale. The three tasks were separated by several-minute pauses. Before the start of the
experiment, audiograms were determined. The whole experiment took about 90 min to complete.
The listeners were tested individually in a soundproof
and highly absorbent listening room (ambient sound level,
<20 dB(A); reverberation time, T60, between 0.25 and
8 kHz, <0.1 s). The experimental sounds were presented
using ambisonic technology (Gerzon, 1973; Poletti, 2005;
Spors et al., 2013) with six loudspeakers placed in a hexagonal configuration surrounding the listener, with all loudspeakers located 1.6 m apart and 1.6 m from the listener at
the center point of the hexagon (Fig. 2, left). The loudspeakers were mounted 1.1 m high to match the approximate
position of the listeners ears when seated (Fig. 2, right). The
sounds were presented at the same levels as recorded, that is,
from 52 to 77 dB LAeq,30s. In the additional experiment (see
Sec. III C), all sounds were set equal to 59 dB LAeq,30s. The
presented levels were calibrated using a sound level meter
(Svan 959, SVANTEK, Warsaw, Poland)
The experiment was controlled by an application programmed in Pd-Extended 0.42.5 (Puckette, 1996), which
instructed Reaper audio digital workstation software to play
the sounds using an internal sound card (RME HDSPe
MADI FX, Haimhausen, Germany) connected to external
MADI to AES/EBU converters (RME ADI-6432R) that fed
the active loudspeakers (Genelec 8130A, Iisalmi, Finland).
The experimental sounds were evaluated on an Android tablet (Samsung Galaxy Note 10.1, Seoul, South Korea) using a
software application developed for this study in JAVA,
Android 4.0.4 (Google, Mountain View, CA). The Pd

FIG. 2. (Color online) Schematic setup of the six-channel ambisonic loudspeaker setup (left). Photo of listener (right).
3046

J. Acoust. Soc. Am. 138 (5), November 2015

application communicated with the other software components over a local network using the OSC protocol.
E. Participants

The main experiment involved 57 listeners (36 females

and 21 males) aged 1954 yr (mean age 27 yr). In an additional experiment, described in Sec. III B, 36 listeners were
tested (17 females and 19 males) aged 1948 yr (mean
age 25 yr). All participants hearing status was tested using
an audiometer (Interacoustics model AD226 diagnostic audiometer, Assens Denmark); the listeners had hearing thresholds below 30 dB hearing level (HL) in their best ear for the
tested frequencies 0.5, 1, 2, 3, 4, and 6 kHz.
III. RESULTS AND DISCUSSION
A. Perceptual results

In the sorting task, the listeners were free to create as

many groups as they found appropriate, and between two
and 16 groups were created (median six groups). The
number of times that two sounds were sorted into the same
group was used as an ordinal measure of their perceived
similarity. The similarity matrix of all sound pairs was subjected to ordinal MDS using the PROXSCAL algorithm
(IBM Statistics SPSS 21), and a two-dimensional solution
(Fig. 3) was found to fit the data well (S-stress values for
one-, two-, and three-dimensional solutions were 0.24, 0.01,
and 0.005, respectively).
Distinct groups of sounds emerged from the MDS solution (Fig. 3). One group of sounds (circled solid) comprised
recordings of fountains with large rising jets or large volumes of downward falling water. The qualitative descriptions given by the listeners and the authors listening
through the sounds suggested that the common perceptual
characteristic of this group of sounds was a loud steady-state
and waterfall-like sound character. At the other end of the

FIG. 3. Two-dimensional multidimensional scaling (MDS) solution.

Symbols refer to three groups of sounds identified from the solution: loud
steady-state sounds (circled solid), moderately loud sounds (open circles),
and soft variable sounds (circled plus). Numbers rank-order the sounds from
most pleasant (1) to least pleasant (32).
Radsten Ekman et al.

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 188.24.122.154 On: Tue, 08 Mar 2016 15:45:56

perceptual space was a set of sounds (circled plus) from

recordings of small fountains with one or a few small rising
jets. Their common perceptual characteristic was a soft variable sound with a purling-rippling sound character. Between
these distinct groups of loud steady-state and soft variable
sounds was a set of sounds (open circles) from various fountains that generated moderately loud sounds, some of a
steady-state character and others with a more variable temporal pattern. The sounds in this intermediate group were
difficult to characterize in terms of common perceptual features, and the listeners used a variety of qualitative descriptors to describe these sounds (that often were sorted in
separate groups by individual listeners). Visual inspection of
the perceptual space (Fig. 3) suggests a cluster of sounds in
the lower-middle part of the space, but listening through
these sounds did not suggest any obvious common characteristics. The most obvious feature of these sounds may simply
be their lack of perceptual characteristics typical of either
the loud steady-state or soft variable sound groups.
After the sorting task, the listeners assessed the sounds
with respect to perceived pleasantness. The numbers in Fig. 3
refer to the rank order the sounds from most to least pleasant,
based on the average pleasantness scores. High numbers (low
pleasantness) are found among the loud steady-state group
(circled solid), whereas low numbers (high pleasantness) are
found among the soft variable group (circled plus). However,
there are exceptions to this rule. For example, sounds 7 and
8, which were among the 10 most pleasant sounds, are
located closer to the loud steady-state than the soft variable
group, whereas the slightly less pleasant sounds 10 and 13
are located firmly in the soft variable group.
B. Acoustic and psychoacoustic analyses

Figure 4 shows time-histories (left diagram) and

narrow-band spectra (right diagram) for a sample of sounds
from each of the three groups identified in the MDS solution
(Fig. 3). The time-histories of the sampled sounds illustrate
well the difference between the three groups of sounds in
terms of overall level and temporal variability: The three
sounds representing the group of soft-variable sounds (#13)

are characterized by lower levels and larger fluctuations than

the sounds repressing the middle group (#1618), which, in
turn, have lower levels and larger variations than the sounds
from the loud steady-state group (#3032). The difference in
overall levels between sounds is also visible by the vertical
position of their spectra (right diagram). The spectral shapes,
however, were not distinctly different across the three groups
of fountain sounds. There was a tendency for relatively more
energy in the high frequency part of the spectrum for the
loud steady-state sounds and moderately loud sounds compared to the soft-variable sounds, however there were also
notable exceptions to this pattern, as discussed below in relation to spectral centroids.
Figure 5 explores how overall level (LAeq,30s), temporal
variability (SDLA) and spectral centroid (SC) was related to
the pleasantness scores. A strong negative relationship was
seen between pleasantness and overall level (leftmost diagram), that is, high levels were associated with low pleasantness scores. The reversed trend was seen for the relationship
between variability and pleasantness (middle diagram), that
is, high variability was associated with high pleasantness.
This relationship was almost as strong as the relations
between overall level, especially in terms of rank-order correlations which were less influenced by the two sounds with
highest variability values (>2 dB, #7 and #4). For spectral
centroid, the relationship was weak (rightmost diagram).
The spectra of the least pleasant sounds, all of type loud
steady-state, had centroids about 3 kHz, but this was also
true for several moderately pleasant sounds. The three most
pleasant sounds, all of type soft variable, had their SC at
slightly lower frequencies, about 2.5 kHz, but pleasant
sounds were also found among those with high SC, notably
sound #4 with a SC of 3.7 kHz. The psychoacoustic indicator
sharpness was highly negatively correlated with pleasantness scores (Pearsons linear coefficient of correlation,
rP 0.71, and Spearmans rank-order coefficient of correlation, rS 0.72). This is not surprising, given the high
negative correlation between pleasantness and overall level,
and the fact that sharpness is not only related to spectral envelope but also to sounds overall level (and thereby

FIG. 4. (Color online) Time-histories: A-weighted SPL (fast) versus time (left panel) and 1/96-octave-band spectra (right panel) for sounds #13 (soft variable), #1618 (moderately loud), and #3032 (loud steady-state).
J. Acoust. Soc. Am. 138 (5), November 2015

Radsten Ekman et al.

3047

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 188.24.122.154 On: Tue, 08 Mar 2016 15:45:56

FIG. 5. Average pleasantness ratings of water-fountain sounds as a function of overall SPL (LAeq,30s, leftmost diagram), standard deviation of A-weighted
instantaneous SPLs, (SDLA, middle diagram), and spectral centroid (SC, rightmost diagram). Statistics and p-values refer to linear (Pearsons rP) and rankorder (Spearmans rS) coefficients of correlation. Symbols identify three groups of sounds (cf. Fig. 3): loud steady-state sounds (circled solid), moderately loud
sounds (open circles) and soft variable sounds (circled plus). Numbers rank-order the sounds from the most pleasant (1) to least pleasant (32).

loudness). Correlations between perceptual attributes and

sharpness are therefore difficult to interpret in experiments
with large SPL variations; the sharpness measure is more
relevant for experiments with equal SPLs as discussed next.
C. Additional equal-SPL experiment

The sounds overall level was strongly related to their

location in the perceptual space and to their pleasantness
scores. However, correlation does not imply causation, and
this is not least true for correlations between acoustic measures and environmental sounds. Listeners may have used
other perceptual characteristics that co-varied with the
sounds overall SPL (LAeq,30s), such as temporal variability
or spectral envelope.
To control for the effect of overall level on perceived
similarity and pleasantness, an additional experiment was
conducted in which the fountain sounds overall SPLs were
set equal to 59 dB LAeq,30s. This procedure left unchanged
the sounds variability (SDLA) and spectral centroid (SC).
The SPL-equalized sounds were assessed using the
same methodology as in the main experiment, but with a
new group of listeners. Data were analyzed in the same way
as in the first experiment described above. A twodimensional MDS solution was again found to fit the
similarity-sorting data well (S-stress values for one-, two-,
and three-dimensional solutions were 0.04, 0.01, and 0.007,
respectively). The solution is shown in Fig. 6. The SPL
equalization did not drastically change the relative location
of the experimental sounds in the perceptual spaces. That is,
the three clusters of sounds identified in the first experiment
(circled plus, open circle, and circled solid) were still discernible in the solution obtained in the equal-SPL experiment. Exceptions include sounds 7 and 24, which moved
from the moderately loud group to the group of fountains
generating loud steady-state sounds, and sound 27, which
moved in the opposite direction. These exceptions notwithstanding, the results suggest that equating the sounds SPLs
did not drastically influence how the sounds were sorted in
terms of perceived similarity.
In contrast, the SPL equalization considerably changed
the relative pleasantness ratings of the sounds. In particular,
3048

J. Acoust. Soc. Am. 138 (5), November 2015

several sounds from the fountains producing soft variable

sounds were now assessed as less pleasant than sounds from
the group of fountains generating moderately loud sounds.
This is seen in the leftmost diagram of Fig. 7, which plots
the pleasantness ratings from the additional equal-SPL
experiment as a function of the corresponding ratings from
the first experiment. The strongly reduced variability in
pleasantness scores in the additional equal-SPL experiment
(y-axis) compared to the first experiment (x-axis) verifies the
conclusion form the first experiment that overall level was
the main determinant of the variability in pleasantness
scores.
Note, however, that there was a moderately strong relationship between the two experiments pleasantness scores
(leftmost diagram of Fig. 7), suggesting that the equal-SPL
procedure did preserve sound characteristics relevant for perceived pleasantness. In particular, several of the sounds from
fountains generating loud steady-state sounds were still

FIG. 6. Two-dimensional multidimensional scaling (MDS) solution from

the additional equal-SPL experiment. Symbols identify three groups of
sounds identified based on the multidimensional scaling solution from the
first experiment (cf. Fig. 3): loud steady-state sounds (circled solid), moderately loud sounds (open circles), and soft variable sounds (circled plus).
Numbers rank-order the sounds from most pleasant (1) to least pleasant
(32), based on the result of the first experiment.
Radsten Ekman et al.

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 188.24.122.154 On: Tue, 08 Mar 2016 15:45:56

FIG. 7. Average pleasantness ratings of water-fountain sounds in the additional equal-SPL experiment as a function of pleasantness ratings from the first
experiment (leftmost diagram), standard deviation of A-weighted instantaneous SPLs (SDLA, middle diagram), and spectral centroid (SC, rightmost diagram).
Statistics and p-values refer to linear (Pearsons rP) and rank-order (Spearmans rS) coefficients of correlation. Symbols identify three groups of sounds identified based on the multidimensional scaling solution from the first experiment (cf. Fig. 3): loud steady-state sounds (circled solid), moderately loud sounds
(open circles), and soft variable sounds (circled plus). Numbers rank-order the sounds from most pleasant (1) to least pleasant (32), based on the result of the
first experiment.

assessed as among the least pleasant sounds. This suggests

that the character of these sounds was inherently unpleasant
independent of their overall level.
The correlation between temporal variability and pleasantness scores was low and not statistically significant. Thus,
eliminating the variability in overall level strongly reduced
the correlation between temporal variability and pleasantness scores observed in the first experiment (cf. Fig. 5, middle diagram). This indicates that the high positive correlation
observed in the first experiment to a significant extent was
confounded by the sounds overall level.
In contrast, the correlation between spectral centroid
and pleasantness scores was moderately high in the equalSPL experiment (rightmost diagram of Fig. 7), with linear
and rank-order correlations of rP 0.38 and rS 0.49,
respectively (excluding sound #20 as a potential outlier
would increase these coefficients to rP 0.53 and

rS 0.60). These correlations were higher than observed in

the first experiment (cf. rightmost diagram Fig. 5), suggesting that spectral envelope may play a role for pleasantness
scores in experiments where overall level does not dominate
the assessments.
To compare with previous studies (e.g., Galbrun and
Ali, 2013), sharpness values were also calculated for the
SPL-equalized sounds. For such sounds, sharpness mainly
captures variation in amount of high frequency content of
sounds. Figure 8 shows the relationships between pleasantness of SPL-equalized sounds and sharpness. The pattern of
data and size of correlations is similar to the corresponding
figure for spectral centroid (cf. Fig. 7, rightmost diagram), as
would be expected from the high correlation between sharpness and SC (rP 0.86). The negative coefficients show that
for this set of sounds, amount of high frequency sounds, as
reflected in high values on SC and sharpness, was associated
with lower pleasantness scores.

IV. GENERAL DISCUSSION

FIG. 8. Average pleasantness ratings of water-fountain sounds in the additional equal-SPL experiment as a function of the psychoacoustic measure
sharpness. Statistics and p-values refer to linear (Pearsons rP) and rankorder (Spearmans rS) coefficients of correlation. Symbols identify three
groups of sounds identified based on the multidimensional scaling solution
from the first experiment (cf. Fig. 3): loud steady-state sounds (circled
solid), moderately loud sounds (open circles), and soft variable sounds
(circled plus). Numbers rank-order the sounds from the most pleasant (1) to
least pleasant (32), based on the result of the first experiment.
J. Acoust. Soc. Am. 138 (5), November 2015

In this study, a set of 32 water-fountain sounds recorded

in urban public spaces were assessed with regard to perceived similarity and pleasantness. The perceptual space
obtained from the similarity sortings suggested perceptually
distinct groups of sounds with different acoustic profiles.
The perceived pleasantness of the water-fountain sounds was
strongly related to their overall SPLs and temporal variability, but weakly related to spectral centroid. However, in the
additional experiment, in which sounds were set equal in
overall SPL, a negative relationship was found between
pleasantness and spectral centroid or sharpness, suggesting
that spectral envelope may influence pleasantness scores in
experiments where overall level does not dominate pleasantness assessments. The additional experiment also demonstrated that the unpleasant sounds from some of the large
fountains remained unpleasant even after equalizing their
SPLs. This suggests that these fountains generated sounds
Radsten Ekman et al.

3049

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 188.24.122.154 On: Tue, 08 Mar 2016 15:45:56

that have an inherently unpleasant sound character independent of their overall SPL.
A. Relation to previous research

The present results agree with those of previous studies

finding that water features in which large amounts of water
impact on water tend to generate unpleasant sounds, for
example, sounds from natural waterfalls (Radsten-Ekman
et al., 2013), waterfall-like fountains (Galbrun and Ali, 2013),
and large jet-and-basin fountains (Axelsson et al., 2014).
Sounds from such structures are characterized by high SPLs,
broadband frequency content and fairly steady-state time histories. In contrast, pleasant water sounds typically have low
SPLs and variable time histories (Watts et al., 2009).
Examples include sounds from natural streams (Galbrun and
Ali, 2013), sea waves (Radsten-Ekman et al., 2013), and
fountains with few and small jests, as examined in the present
study.
In the first experiment of the present study, there was no
strong relationship between pleasantness scores and the
sounds spectral centroid. This stand in contrast to previous
studies on water-generated sounds (Watts et al., 2009; Jeon
et al., 2012; Galbrun and Ali, 2013), which have reported
relationships between preference-related attributes and
parameters related to the sounds spectral envelope (including the psychoacoustic indicator sharpness, discussed further
below). This discrepancy may partially be related to the
limited variation in spectral envelopes of the present experiments sounds (cf. Fig. 4) compared to previous studies. For
example, two previous studies used recordings of small
water features with a variety of materials, including water,
concrete, stones, and gravel, on which the running water
impacted. This variation in impact material causes large
between-sound variation in spectral composition (Galbrun
and Ali, 2013), which could explain the stronger relationship
between preference and spectral envelope than found in the
present study of fountains, in all of which water was the
impact material. However, a more important factor is probably the limited variation in SPLs in the cited studies
compared to the first experiment of the present study. For
example, Galbrun and Ali (2013) used water generated
sounds normalized to 55 dB LAeq, Watts et al. (2009) conducted separate experiments, each with a constant SPL
(ranging from 43 to 60 dB LAeq across experiments), and
Jeon et al. (2012) mixed road-traffic noise with watergenerated sounds normalized to 52 dB LAeq in one condition
and 72 dB LAeq in another condition. It is possible that spectral aspects mainly influences preference ratings under such
equal-SPL conditions, as suggested by the much stronger
correlation between pleasantness and spectral envelope in
the equal-SPL experiment of this study (all sounds set to
59 dB LAeq) compared to the first experiment (SPLs from 52
to 77 dB LAeq). A large variation in SPLs implies a large
variation in perceived loudness, the main determinant of
perceived noise annoyance (e.g., Berglund et al., 1990). The
present results are consistent with the notion that loudness
also is a main determinant of unpleasantpleasantness of
water-generated sounds, and that a large loudness-variation
3050

J. Acoust. Soc. Am. 138 (5), November 2015

(as in the first experiment of the present study) may dominate the perception to the extent that less salient variation in
spectral envelope has little influence on pleasantness assessments. From a basic research perspective, it may thus be
advisable to restrict the variation in overall SPL in listening
experiments to explore spectral predictors of preferences for
water generated sounds, as in this studys equal-SPL experiment. From an applied perspective, it is of course more
relevant to present sounds at realistic SPLs, as in this studys
first experiment.
Several of the studies cited above reported relationships
between preference-related attributes of water-generated
sounds and the psychoacoustic indicator sharpness. As already
mentioned, these studies used water-generated sounds of
approximately equal SPLs. For such sounds, sharpness is
mainly a measure of spectral envelope, as was illustrated by
the high correlation between sharpness and spectral centroid
(rP 0.86) of sounds in the present studys equal-SPL experiment. In that experiment, sharpness (and spectral centroid)
was negatively associated with perceived pleasantness. This
agrees with listening studies of Galburn and colleges (Galbrun
and Ali, 2013; Galbrun and Calarco, 2014) who used various
water-generated sounds, including waterfall, fountain, and
stream sounds. In contrast, Watts et al. (2009) reported a positive correlation between preference and sharpness. Galburn
and Ali speculate that this inconsistency may be because they
used both upward and downward flows, whereas Watts et al.
(2009) used downward flows only, including low-sharpness
sounds that might have evoked negative associations of water
running down drains. Galbrun and Ali (2013) suggests that
sharpness might not be a key factor driving preferences for all
types of water features, whereas temporal variation might be,
in line with their finding of a positive relationship between
temporal variability and preference. This is an interesting idea
that should be explored further. It agrees with the positive
association between pleasantness and temporal variability in
the first experiment of the present study. However, it remains
to be seen whether temporal variability was a causal factor or
just a covariate, because the correlation between pleasantness
and temporal variability was much reduced (and non-significant) in the additional experiment using SPL-equalized sounds.
In fact, spectral envelope measures (SC and sharpness) were
stronger related to pleasantness scores than temporal variability
in the equal-SPL experiment. Jeon et al. (2012) used recordings of fountains in public open spaces mixed with road traffic
noise and found a negative correlation between sharpness and
calmness, in line with Galbrun and Ali (2013) and in line with
the present results from the equal-SPL experiment. However,
Jeon et al. (2012) also reported a positive correlation between
sharpness and preference ratings of the same sounds. More
research is clearly needed to clarify the role of spectral factors
for preference ratings of water-generated sounds.
B. Implications

The whole idea of fountains is, of course, to enhance the

quality of the spaces where they are erected. Several of the
large fountains included in the present study were visually
attractive and obviously designed to produce an interesting
Radsten Ekman et al.

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 188.24.122.154 On: Tue, 08 Mar 2016 15:45:56

and aesthetically attractive interplay between fountain

structure and water movement (see photos in Fig. 1).
Unfortunately, the results of the present study suggest that
several of these fountains generated unpleasant sounds,
which probably detracts from the fountains overall contribution to the spaces where they are installed. These results are
based on correlations and do not allow causal inferences.
This said, decreased flow rate would seem to be a good start
when seeking to improve the sound quality of problematic
fountains, because this would reduce loudness and increase
sound variability. At the same time, the soft and pleasant
purling-rippling sound of the small fountains examined here
would hardly be appropriate for a large fountain installed in
a vibrant or noisy urban space, where its sound could be
masked. Future research is challenged to consider the design
of large fountains that generate pleasant sounds that at the
same time are congruent with their size, structure, and
dynamics as well as their locations.
C. Strengths and limitations

A strength of this study is that it used recordings from a

large set of real fountains in urban public spaces. Though by
no means a random selection of fountains in the Stockholm
region, the set included a large variety of fountain types and
represents the types of fountain sounds one may encounter in
Stockholm or a similar modern city. The recordings were
made near the fountains, so the results cannot be generalized
to fountain sounds heard at greater distances where they
would be heard mixed with other prominent sounds in the
area. The present experiments were restricted to sounds in
the absence of visual information. Audiovisual interactions
may influence the results of listening experiments. However,
the effects of visual information on auditory perception are
small in most cases (Nilsson et al., 2014), although visual information strongly influences the overall assessment of a
place (Hong and Jeon, 2013; Galbrun and Calarco, 2014).
The experiments used advanced audio reproduction based on
ambisonic technology, which allows for very realistic loudspeaker presentation of soundscapes. A novelty of the present experiments compared with previous studies of watergenerated sound is that a sorting methodology was used that
does not require predefined definition of the perceptual
attributes to which the listener should attend. In the present
application, the results suggested distinct perceptual difference between soft variable and loud steady-state sounds
from water fountains.
D. Conclusions

The following conclusions can be drawn from the

results of the present study:
(1) Distinct groups of water-fountain sounds were identified: soft variable sounds from small fountains and loud
steady-state sounds from larger fountains. Acoustically,
the soft variable sounds were characterized by low
overall levels and high temporal variability, whereas
the opposite pattern characterized steady-state sounds.
A third group of moderately loud fountain sounds
J. Acoust. Soc. Am. 138 (5), November 2015

included both sounds with high and low temporal

variability.
(2) Spectral envelope may play a role for the pleasantness of
fountain-generated sounds, but mainly for assessments
of sounds of similar overall levels, for which a negative
correlation between the sounds pleasantness and their
spectral centroid or sharpness was suggested.
(3) High flow-rate fountains generating steady-state sounds
seem to have an inherently unpleasant sound character
irrespective of their overall SPL and, thereby, their loudness. From a soundscape perceptive, it may be advisable
to avoid fountain designs that generate such sounds.
ACKNOWLEDGMENTS

This research was conducted in the Sound Cities

research program, funded by the Marianne and Marcus
Wallenberg foundation.
Aures, W. (1985). Berechnungsverfahren f
ur den sensorischen wohlklang
beliebiger schallsignale (Calculation methods for perceived quality of
sound signals), Acustica 59, 130141.
Nilsson, M. E., and Berglund, B. (2010). A principal compoAxelsson, O.,
nents model of soundscape perception, J. Acoust. Soc. Am. 128,
28362846.
Nilsson, M. E., Hellstr
Axelsson, O.,
om, B., and Lunden, P. (2014). A field
experiment on the impact of sounds from a jet-and-basin fountain on
soundscape quality in an urban park, Landscape Urban Plann. 123,
4960.
Berglund, B., Preis, A., and Rankin, K. (1990). Relationship between loudness and annoyance for ten community sounds, Environ. Int. 16, 523531.
Coxon, A. P. M. (1999). Sorting Data: Collection and Analysis (Sage,
London), pp. 198.
De Coensel, B. D. C. B., Vanwetswinkel, S., and Botteldooren, D. (2011).
Effects of natural sounds on the perception of road traffic noise,
J. Acoust. Soc. Am. 129, EL148EL153.
Dramstad, W. E., Tveit, M. S., Fjellstad, W., and Fry, G. L. (2006).
Relationships between visual landscape preferences and map-based
indicators of landscape structure, Landscape Urban Plann. 78,
465474.
Galbrun, L., and Ali, T. T. (2013). Acoustical and perceptual assessment of
water sounds and their use over road traffic noise, J. Acoust. Soc. Am.
133, 227237.
Galbrun, L., and Calarco, F. M. (2014). Audio-visual interaction and perceptual assessment of water features used over road traffic noise,
J. Acoust. Soc. Am. 136, 26092620.
Gerzon, M. A. (1973). Periphony: With-height sound reproduction,
J. Audio Eng. Soc. 21, 210.
Gygi, B., Kidd, G. R., and Watson, C. S. (2007). Similarity and categorization of environmental sounds, Percept. Psychophys. 69,
839855.
Hong, J. Y., and Jeon, J. Y. (2013). Designing sound and visual components for enhancement of urban soundscapes, J. Acoust. Soc. Am. 134,
20262036.
Jeon, J. Y., Lee, P. J., You, J., and Kang, J. (2010). Perceptual assessment
of quality of urban soundscapes with combined noise sources and water
sounds, J. Acoust. Soc. Am. 127, 13571366.
Jeon, J. Y., Lee, P. J., You, J., and Kang, J. (2012). Acoustical characteristics of water sounds for soundscape enhancement in urban open spaces,
J. Acoust. Soc. Am. 131, 21012109.
Kuppens, P., Tuerlinckx, F., Russell, J. A., and Barrett, L. F. (2013). The
relation between valence and arousal in subjective experience, Psychol.
Bull. 139, 917940.
Nasar, J. L., and Li, M. H. (2004). Landscape mirror: The attractiveness of
reflecting water, Landscape Urban Plann. 66, 233238.
Nilsson, M. E., Alvarsson, J., Radsten-Ekman, M., and Bolin, K. (2010).
Auditory masking of wanted and unwanted sounds in a city park, Noise
Control Eng. J. 58, 524531.
Radsten Ekman et al.

3051

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 188.24.122.154 On: Tue, 08 Mar 2016 15:45:56

Nilsson, M. E., Botteldooren, D., Jeon, J. Y., Radsten-Ekman, M., De

Coensel, B., Joo, Y., Maillard, J., and Vincent, B. (2014). Perceptual
effects of noise mitigation, in Environmental Methods for Transport
Noise Reduction, edited by M. E. Nilsson, R. Klbo, and J. Bengtsson
(Taylor & Francis, New York), pp. 195219.
Poletti, M. A. (2005). Three-dimensional surround sound systems based on
spherical harmonics, J. Audio Eng. Soc. 53, 10041025.
Puckette, M. (1996). Pure data: Another integrated computer music environment, in Proceedings of the Second Intercollege Computer Music
Concerts, pp. 3741.

Radsten-Ekman, M., Axelsson, O.,

and Nilsson, M. E. (2013).
Effects of sounds from water on perception of acoustic environments dominated by road traffic noise, Acta Acust. Acust. 99,
218225.
Russel, J. A. (1980). A circumplex model of affect, J. Pers. Soc. Psychol.
39, 11611178.

3052

J. Acoust. Soc. Am. 138 (5), November 2015

Russel, J. A., and Mehrabian, A. (1978). Approach-avoidance and affiliation as functions of the emotion-eliciting quality of an environment,
Environ. Behav. 10, 355387.
Spors, S., Wierstorf, H., Raake, A., Melchior, F., Frank, M., and Zotter, F.
(2013). Spatial sound with loudspeakers and its perception: A review of
the current state, Proc. IEEE 101, 19201938.
Vastfjall, D., Gulbol, M.-A., Kleiner, M., and Garling, T. (2002). Affective
evaluations of and reactions to exterior and interior vehicle auditory quality, J. Sound Vib. 255, 510518.
Watts, G. R., Pheasant, R. J., Horoshenkov, K. V., and Ragonesi, L. (2009).
Measurement and subjective assessment of water generated sounds,
Acta Acust. Acust. 95, 10321039.
White, M., Smith, A., Humphryes, K., Pahl, S., Snelling, D., and Depledge,
M. (2010). Blue space: The importance of water for preference, affect,
and restorativeness ratings of natural and built scenes, J. Environ.
Psychol. 30, 482493.

Radsten Ekman et al.

Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 188.24.122.154 On: Tue, 08 Mar 2016 15:45:56