Applied Acoustics 92 (2015) 4760

Contents lists available at ScienceDirect

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

Psychoacoustic, physical, and perceptual features of restaurants: A eld

survey in Singapore
PerMagnus Lindborg
Nanyang Technological University, Singapore
KTH Royal Institute of Technology, Stockholm, Sweden

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

Article history: Sound is a multi-faceted phenomenon and a critical modality in all kinds of sevicescapes. At restaurants,
Received 20 June 2014 our senses are intensively stimulated. They are social places that depend on acoustic design for their suc-
Received in revised form 20 October 2014 cess. Considering the large economic interests, surprisingly little empirical research on the psychoacous-
Accepted 5 January 2015
tics of restaurants is available. Contributing to theory building, this article proposes a typology of
Available online 30 January 2015
designed and non-designed sonic elements in restaurants. Results from a survey of 112 restaurants in
Singapore are presented, with a focus on one element of the typology, namely interior design materials.
Keywords:
The collected data included on-site sound level, audio recordings from which psychoacoustic descriptors
Soundscapes
Environmental quality
such as Loudness and Sharpness were calculated, perceptual ratings using the Swedish Soundscape Quality
Psychoacoustics protocol, and annotations of physical features such as Occupancy. We have introduced a measure,
Perceptual features Priciness, to compare menu cost levels between the surveyed restaurants. Correlation analysis revealed
Interior design several patterns: for example, that Priciness was negatively correlated with Loudness. Analysis of annota-
Restaurants tions of interior design materials supported a classication of the restaurants in categories of Design Style
and Food Style. These were investigated with MANOVA, revealing signicant differences in psychoacou-
stic, physical, and perceptual features between categories among the surveyed restaurants: for example,
that restaurants serving Chinese food had the highest prevalence of stone materials, and that Western-
menu places were the least loud. Some implications for managers, acoustic designers, and researchers
are discussed.
2015 Elsevier Ltd. All rights reserved.

1. Introduction shops, commuter areas, lounges, hospitals and so forth. Research

into servicescapes considers the multiple factors that affect
The introduction reviews articles about restaurant soundscapes peoples emotions, responses, and behaviours when engaged in
published in the recent literature on acoustics and hospitality hedonic interaction in such environments. Designable physical
research, with a focus on those that report acoustic measures. A factors in servicescapes include temperature, lighting, noise, music,
typology for acoustic design elements in restaurants is then and scent. Bitner suggested that the ambient conditions affect the
proposed. ve senses in purely physiological ways. Noise that is too loud
may cause physical discomfort, the temperature of a room may
cause people to shiver or perspire, the air quality may make it dif-
1.1. Servicescapes and soundscapes cult to breathe, and the glare of lighting may decrease ability to
see and cause physical pain. All of those physical responses may
The concept of servicescape was introduced by Mary Jo Bitner, in turn directly inuence whether or not people stay in and enjoy
referring to the built environment. . . the manmade, physical sur- a particular environment. (ibid. p. 64)
roundings as opposed to the natural or social environment [6, Sound is a critical modality of servicescapes. Since the 1970s, R.
p. 58]. A more narrow denition has since then emerged, pointing Murray Schafer used the term soundscape to refer to all aural
specically towards hospitality service settings, such as restaurants, elds of study scientic and creative (often interdisciplinary)
such as soundscape design, acoustic ecology, sound walks, and so
Corresponding author. Address: School of Art Design Media, 81 Nanyang Drive, forth [42,43,45]. As Berglund et al. [5] pointed out, soundscapes
Nanyang Technological University, 637458 Singapore. Tel.: +65 6316 8727; fax: +65 are complex perceptual phenomena. It is possible to simulta-
6795 3140.
neously have positive and negative perceptions of a sonic environ-
E-mail address: permagnus@ntu.edu.sg

http://dx.doi.org/10.1016/j.apacoust.2015.01.002
0003-682X/ 2015 Elsevier Ltd. All rights reserved.
48 P. Lindborg / Applied Acoustics 92 (2015) 4760

ment, or rather, aspects of it. Perception depends on expectations,  Investigating ten bars, cafs and restaurants in Wellington,
preferences, mood, and current activities. While we react physio- Christie and Bell-Booth [11] found that the highest sound level,
logically to some aspects of soundscapes in ways determined by in a restaurant, was = 81 dB.
our biological setup [25], other aspects are dealt with in cognitive,  Rusnock and Bush [40] surveyed 30 bars and restaurants in
individual ways, yet predictably moderated by personality traits Florida and recorded 1520 random instantaneous sound level
[26]. readings (ibid. p. 110) at each place. They found eight with
maximum levels above 85 dBA and practically all the rest in
1.2. Restaurants the range between 75 and 85 dBA .
 Frid [18] included multiple acoustic measures in her investiga-
In Bitners classication scheme, restaurants are characterised tion of 31 restaurants in Stockholm (cafs, lunch restaurants,
as offering interpersonal services. . . of elaborate physical com- and fast-food places). She found sound levels in the range
plexity. She stressed the importance of considering how environ- LAeq = {62.7 . . . 82.4} dB, and the average across all locations
mental factors affect the well-being of employees as well as was 73.7 dBA.
customers, and found that managers might control the physical
surroundings but frequently lacked an understanding of their A detailed comparison between these results is difcult since
impact on people. Her call for wider application of environmental the methods used were not always reported and might have been
psychology was echoed by Novak et al. [36], who claimed that hos- different. Peak sound level has little informative value on its own,
pitality researchers have paid too little attention to how the phys- though together with LAeq and range may indicate the distribution
ical environment of servicescapes inuences the customers of sound levels. There are anecdotal reports of average noise levels
emotional responses. The empirical evaluation of soundscape qual- in contemporary restaurants and cafs at 80 dBA or higher, accom-
ity in restaurants and how it affects the overall experience of panied by claims that generally rising volume levels are due to
patrons has only recently started to be addressed. managers seeking a hip and boisterous ambiance [1]; see also
Restaurants are intensively social places. Their design is subject [28]. Recall that Schafer and his collaborators documented a
to competing requirements. They need to attract customers; and so long-term trend of SPL increase of 35 dB per decade in trafc
they have large windows towards the street or open doors towards noise levels and sirens ([42, p. 1856]). Is the same thing happen-
the shopping mall in which they are located, leaking in external ing in restaurants? The published data do not rule out a long-term
noise. They need to give an impression of cleanliness; so oors increase in restaurants, but the records are too sparse to rmly
and tabletop materials are hard and acoustically reective. support this assumption.
Managers and waiters need to scan the tables; so there are few wall It is appropriate to recall the texts of law concerning workplace
partitions to absorb sound and provide lateral rst reec- safety and health. For example, the regulations in Singapore and in
tions. Sometimes, customers want a vibrant atmosphere where USA, which are practically identical in this regard, state that the
appropriate music mixes with voices into an agreeable din; so there employer must take certain actions when there are indications of
is a rationale for reverberant acoustics (cf. metabolic environ- an employees sound exposure exceeding LAeq8h = 85 dB [37,31].
ments, discussed in [20,21]. Other times, customers seek calm to In Europe, the limit is lower, 80 dB [16]. As sound levels go up, reg-
digest the beautiful food or privacy for a spoken conversation. ulation limits go down. Such limits are minimal requirements that
Whatever the expectations are, restaurateurs must get return on some audiologists consider insufcient. In Singapore, Noise-
their investment, and therefore depend on their customers being Induced Deafness (NID) has been the most prevalent occupational
overall pleased with the experience; so that they may appraise pos- disease for many years, accounting for 80% of cases, and with sig-
itively and display return behaviour (cf. [33,36]). nicant year-on-year increases. While ministries have shown a
Considering the large economic interests that depend on suc- willingness to stimulate awareness and engagement with the pri-
cessful restaurant design, and Bitners constructive call twenty vate sector, empirical acoustic measurements are not part the pol-
years ago for more attention to multi-sensorial responses to ser- icy publications (ibid; also [30]). One important aim of the present
vicescapes, including sound, there are surprisingly few published work has been to provide data from a fairly large sample of
studies that report objective acoustics. For a long time the basic restaurants.
measurement (often the only one considered) has been Sound
Pressure Level (SPL, in dB re 20 lPa), typically A-weighted and 1.3. Acoustic design
averaged over time (e.g. LAeq3m). Most researchers agree that this
descriptor on its own is insufcient to predict perceptual responses Building on earlier research [6,32,39,20,15,33], we have drafted
to complex sonic environments, and several others that aim to a typology of acoustic design elements that have a bearing on res-
describe amplitude variation and aspects of timbre have been pro- taurant soundscapes. It has provided a theoretical framework to
posed (cf. [34,19,17]). Nevertheless, LAeq remains a fundamental guide our ongoing investigations. Firstly, a distinction was made
descriptor, and provides a way for comparing sonic environments between non-designed sound sources and designed physical ele-
historically. In our review, we have identied six articles that ments and sources. Sounds within the non-designed category
report on-site sound levels in restaurants: are those beyond the awareness and/or control of restaurant man-
agers and interior designers. They were classied according to ori-
 Lebo et al. [24] recorded average noise levels in 27 restaurants gin: technological, human, or natural [34], and furthermore, as
around San Francisco in the range LAeq = {59 . . . 80} dB, with intermittent or continuous. Within the designed category, distinc-
peaks (noise crests) at 87 dBA in some of the restaurants tions were made between maskers (acoustically active, sound-
(ibid. p. 46). emitting sources though not perceived as such), enclosures, and
 A study by Sweetow and Tate (2000; quoted in [36] reported attenuators (acoustically passive, physical elements). Secondly,
exposure levels of LAeq8h = {50.5 . . . 90} in ve restaurants each type of element was considered as located either inside or
ranging from a noisy bistro to a quiet restaurant/bar (ibid., outside the restaurant itself.
p. 199) at unspecied locations. Table 1 charts the typology, with examples of typical elements.
 Rohrmann [39] took acoustic measures in twelve cafs and res- In a real restaurant, the soundscape might be complex and the dif-
taurants in Melbourne and found levels to be in the range ferent design categories not easily delimited. For example, it could
LAeq1. . .3m) = {56 . . . 85} dB. be argued that chairs scraping the oor is a source of technological
 P. Lindborg / Applied Acoustics 92 (2015) 4760 49

Table 1
A typology of acoustic design elements in restaurants.

Intention Source Category Inside From outside

Non-designed Technological Broad (spatially and temporally) HVACs, fans Trafc, railroads, industry
Point-like (spatially and temporally) Coffee grinders, dish washers Car horns, aircraft
Human Voices Conversation, laughter, singing
Handled objects Cutleries, chairs scraping the oor Construction work
Natural Biophonies Caged birds Birds, insects
Geophonies Wind, rain, waves, streams, thunder
Designed Active Music Moozak, live music, biophony simulacra Loudspeakers
e.g. recorded birdsong
Informational maskers Pink noise Geophony simulacra, e.g. fountains
Passive Enclosure materials Walls, oors, ceiling Built-up sound barriers, mounds,
Interior design materials Tablecloth, curtains, carpets greenery, forests

origin, or that caged birds are elements of musical design, i.e. an 2. Survey
imitation of nature rather than nature itself.
Section 2.4 of the present article focusses on interior design The survey was designed to collect recordings and acoustic
materials, which fall under designed, passive in the typology. measurements from a large number of restaurants, together with
In parallel, other ongoing work focus on non-designed, technolog- annotations about physical features and perceptual quality ratings.
ical, point-like elements, and designed, active, music elements.
Bitner [6] pointed out that certain materials can communicate 2.1. Procedure
cues to establish the character of a restaurant. Besides creating an
overall aesthetic impression, the quality of interior design materi- The author and his students individually visited 118 restaurants
als have symbolic meaning. Restaurant managers know that hard between 17 August and 9 December 2012. The author collected
oors signal cleanliness, and thick curtains intimacy. White table data from 20 locations, using an ANSI Type 2 compliant and cali-
cloths symbolise service and higher prices, and plastic furnishings brated Sound Pressure Level (SPL) meter (Extech 407790) to record
symbolise casual dining and lower prices. Christie and Bell-Booth un-weighted integrated SPL, i.e. LZeq. Simultaneously, uncom-
[11] noted that the nishings in practically all their surveyed pre- pressed and un-ltered audio recordings were made using a
mises were acoustically reective. Rusnock and Bush [40] high-quality microphone (CoreSound Tetramic to SoundDevices
remarked that contemporary restaurants typically have an open recorder) and low-cost equipment (Edirol R-09) in parallel. A-
layout, causing noise from the kitchen and bar to spill over into and C-weighted SPL were obtained computationally from the audio
the dining area. They also noted a prevalence for bare surfaces, recordings (W omni channel) using the program Psysound3 [8,9],
with few textiles such as carpets, draperies, or linens. Their gen- taking the corresponding LZeq as the calibration level for each case.
eral impression was that few restaurants made efforts to dampen Students collected data from 98 locations (5 each). They were
noise levels with interior design materials. Frid [18] annotated free to choose any ve or so places, as long as their selection was
design materials in her survey and found that glass walls and reasonably varied in terms of price, style, and location. At least
facades were common, as were tile oorings. Wooden materials one had to be a table-service dining restaurant. The ambient noise
were more often used in cafs and lunch restaurants than in level was measured on-site with a method similar to that
fast-food places. Moreover, cafs were associated with textiles, described by [40, p. 110]. Over a period of one minute, ten readings
while only hard materials and hard chairs were used in the fast- on a low-cost SPL meter (Galaxy Audio CM-130 Check Mate) were
food places. recorded at approximately regular intervals, with Slow response
setting and A-lter weighting. This was followed by another set
of ten readings using the C-lter weighting. The dB-averages of
1.4. Aims these readings were taken as estimates of LAeq1m and LCeq1m,
respectively. Simultaneously, an audio recording was made with
To sum up, servicescapes are complex physical environments, an Edirol R-09.
created to stimulate our multiple senses. While there is some A separate test of the reliability of the estimates obtained by
research on sound levels, qualitative assessments of restaurant this method was made in studio. Recordings of the 20 restaurants
soundscapes and their inuence on the experience have not been collected by the author were played back using high-quality loud-
made. The effect of materials on room acoustics is known, but speakers (Genelec 8030, 7050A) at the LZeq level recorded on-site.
the actual prevalence of these materials in restaurant design is not. He then estimated LAeq1m and LCeq1m with the method described
This situation urged us to investigate restaurants in Singapore, above. The results suggested that LAeq estimates were on average
asking: How important is the soundscape to the experience of 1.5 dB lower than the reference LAeq (calculated as in Sec-
the whole environment? Are prices and noise levels related? Con- tion 2.2.2), while LCeq estimates were typically less than 1 dB
sidering their acoustic impact, which interior design materials are too low. These discrepancies were considered acceptable for the
common? What differences are there between various types of survey. Furthermore, the mean difference between (on-site) LZeq
places: for example, between fast-food and table-service restau- and (calculated) LCeq across the 20 restaurants was 1.0 dB, repre-
rants, between cheap and expensive places, or between places senting the amount of acoustic energy attenuated by the C lter.
serving Western and Asian food? We performed a broad survey Therefore, LZeq estimates for the student-collected locations were
aiming to shed light on these questions, and to provide empirical obtained by taking (on-site) LCeq1m + 1.0. These values were used
data to serve researchers in further work. in the calculation of psychoacoustic descriptors (see Section 2.2.2).
Section 2 of this article describes the procedure of the survey, The method allowed us to use low-cost equipment in this project.
with results and analyses that are briey discussed. Section 3 is a While it can be improved upon, we believe that the estimates are
conclusion of the main ndings. acceptable within the context of an exploratory survey.
50 P. Lindborg / Applied Acoustics 92 (2015) 4760

Each collector lled out two protocols on site: one for annota- rainfall, would be a prominent reason for acoustic differences
tions and one for perceptual ratings. The annotations included between indoor and outdoor locations. In the present data, there
observations about the following physical features: was only one location were the presence of rain was noted (driz-
zle). Since the rain was barely audible in the recording it was
 Location [Name, Address, Date, Time]. deemed reasonable to include this location. Anecdotally, it could
 Situation [Indoors/Outdoors/Comment (unusual location etc)/ be noted that the torrential rainfall in the tropics is sometimes so
Weather (e.g. rain)]. heavy that its impact on the tiled roofs of foodcourts makes spoken
 Food Style [Chinese/Indian/Japanese/Other Asian (specify)/Ital- conversation impossible.
ian/American/Other Western (specify)]. Food Style and Design Style categorisations are discussed in
 Design Style [Restaurant/Coffee house/Canteen/Hawker/Bar/ Section 2.3.
Buffet/Other (specify)].
 Prices [Average set meal (or typical menu)/Coca Cola/Alterna- 2.2.2. Psychoacoustic descriptors
tive typical item & price (specify)]. From the audio recordings were calculated a set of psychoacou-
 Physical Environment [Number of chairs or stools (i.e. estimated stic descriptors [17]; cf. [41]), [19]:
maximum capacity)/Number of people (currently at location &
within sight)/Architecture & materials (free-form)].  Loudness (N, sone) expresses the experience of a sounds inten-
sity. It is related to acoustic intensity and the distribution of
For perceptual ratings we used the Swedish Soundscape Quality energy in the frequency domain.
Protocol (SSQP; [4], which employs 5-point Likert scales anchored  Sharpness (S, acum) is related to the centroid of the sounds fre-
by adjectives (e.g. from disagree completely to agree com- quency magnitude spectrum, and thus to pitch. Loudness has lit-
pletely, from very good to very bad, and so forth) covering tle inuence on sharpness.
four aspects of soundscape perception:  Roughness (R, asper) is related to dissonance, and is mainly
determined by the prevalence of relatively rapid amplitude
 Sonic Quality [Overall, how would you describe the present sur- variations. The perception of roughness reaches a peak for mod-
rounding sound environment?]. ulation frequencies around 70 Hz. Roughness is normally
 Visual Quality [Overall, how would you describe the present ascribed a negative correlation with pleasantness [13].
surrounding visual environment?].  Fluctuation strength (F, magnitude; see below) is related to
 Sound Source [To what extent do you presently hear the follow- relatively slow audible amplitude variations. The perception of
ing 5 types of sounds? (trafc noise, fan noise, other noise, uctuation strength reaches a peak for modulation frequencies
sounds from human beings, natural sounds)]. around 4 Hz. The fact that this rate is similar to that of syllables
 Qualia [To what extent do you agree with the 8 statements in normal speech is believed to result from biological co-adap-
below on how you experience the present surrounding sound tation of speech and hearing functions [13].
environment?. . . pleasant, exciting, eventful, chaotic, annoying,
monotonous, uneventful, calm]. The label qualia does not The program Psysound3 [8,9] was used to calculate the rst
appear in the original publication and is used here for three descriptors. Loudness and Sharpness were given by the
convenience. Dynamic Loudness Model by Chalupper & Fastl [10], with values
sampled every 0.002 s. The algorithm from Daniel and Weber
Six locations were discarded due to missing SSQP ratings (4 out [14] produced Roughness, with values sampled every 0.186 s (xed
of 118) or incorrect recording settings (2). One missing recording by the program).
(Edirol battery failure) collected by the author was replaced by a As for uctuation strength, the model given by Fastl and Zwick-
stereo reduction of the corresponding ambisonic recording. Apart er [17] refers to pure tones and assumes only one modulation fre-
from this single case, all analysed audio was recorded with identi- quency. It is therefore not generally applicable to real-world
cal equipment and settings to produce recordings of homogenous sounds (Densil Cabrera, personal communication, Feb. 2014). At
and reasonably high quality. From each location, the author the time of the analysis, the Psysound3 implementation of uctu-
selected a representative excerpt of 60 s duration, free of any audi- ation strength was being revised. We therefore calculated an esti-
ble artefact (e.g. microphone handling noise). In nine cases, mate using the miructuation function of the MIR Toolbox ([23]).
excerpts between 58 and 30 s were accepted. Its algorithm is similar, except that the excitation levels in Bark
bands are expressed in dB rather than sone. Summing across
2.2. Results bands, Fluctuation strength is therefore expressed as a relative
magnitude, rather than in the Zwicker unit vacil. Values were sam-
The collection process produced a random sample of 112 Sin- pled every 0.25 s, in overlapping windows of 1 s duration.
gaporean restaurants. The methods for processing the data are out- Finally, to facilitate comparison with other studies, the estimate
lined in this section, with results given in tables. The analysis was for A-weighted sound level, obtained on-site, was included (cf.
made in three parts: rst, by pair-wise correlations (Section 2.3); [41]).
then, by restaurant categorisations and annotations of design From the time-series envelopes, statistics for the 10th, 50th, and
materials (Sections 2.5 and 2.6); and last, by multiple analysis of 90th percentiles were calculated. For example, N50 is the median
variance between categories (Section 2.7). loudness, N10 the loudness exceeded 10% of the time (representing
loud events) and N90 the loudness exceeded 90% of the time
2.2.1. Location and categorisations (background). The difference between 10th and 90th percentiles
In regards to Location, analysis of the Time of visit showed that indicates the variability in a descriptor, and thus N1090 indicates
no weekday was over-represented (X2(6) = 5.12, p = 0.53 n.s.), and Loudness Variability (cf. [3]).
that 90% of the observations made between 8 and 22 h. Unsurpris- Table 2a lists results for the psychoacoustic descriptors.
ingly, the distribution showed modes at 13 and 19 h, corre-
sponding to lunch and dinner. 2.2.3. Prices and physical environment
In regards to Situation, more places were indoors (82) than The survey collectors annotated menu prices in Singapore dol-
outdoors (X2(1) = 24, p = 0.00). Weather conditions, in particular lars for Average set meal (102 restaurants out of 112) and Coca
 P. Lindborg / Applied Acoustics 92 (2015) 4760 51

Cola (90), as well as an Alternative typical item (33). The last Each distribution was tested for normality using Shapiro-Wilks
type included many various foods and drinks, such as espresso test. Pleasantness and Eventfulness were normally distributed, but
and the local specialities chicken rice or ice kachang. In order psychoacoustic descriptors, Priciness, and Occupancy were posi-
to make comparisons between these, the perceived costliness of tively skewed. These variables were subjected to power transfor-
each item was estimated on a 5-point Likert scale (from cheap mation using a Box-Cox method implemented in the R package
to expensive) by ve independent raters with local knowledge; AID [12]. The transformed variables were approximately normal
their agreement was high (Cronbachs alpha = 0.89, mean pair- (Shapiro-Wilks W = {0.98 . . . 0.99}, p = {0.053 . . . 0.91}), except
wise correlation = 0.64). The three price estimates (set meal, S50, which could not be successfully transformed with this
coca-cola, alternative) were z-transformed separately and then method. This was also the case for the one-dimensional SSQP rat-
averaged, yielding a variable named Priciness, describing the rela- ings Sonic Quality, Visual Quality, and the ve Sound Source scales.
tive cost levels among the surveyed restaurants. This measure is Pair-wise correlations were then calculated, with Pearsons
assumed to be important to the perception of how expensive a product-moment correlation r for comparisons where both vari-
place is. ables had normal distributions, and Kendalls tau for all others.
The median number of chairs at each location was 100%, with Kendalls tau is a rank-based correlation statistic with known stan-
90% of the locations having between 30 and 500 chairs. The median dard error.
number of guests was 35%, with 90% of the restaurants having Table 3 shows the cross correlation matrix, followed by a brief
between 5 and 100 guests. Both of these distributions were posi- discussion of relationships of interest.
tively skewed. In line with earlier research (e.g. [24], Occupancy
was calculated as the reciprocal between the number of guests 2.3.1. Psychoacoustic descriptors
and the number of chairs. Occupancy is assumed to be important The correlation between LAeq and Loudness (N50) was high
to the perception of how crowded a place is. (r = 0.82) and there were only minor differences in how they
Table 2b includes results for Priciness and Occupancy. correlated with other variables. For practical purposes, LAeq might
thus be an adequate descriptor of loudness (cf. [41]). Loudness cor-
2.2.4. Sound sources and ratings of quality related strongly with Roughness (r = 0.71) and Loudness Vari-
The SSQP ratings of Sonic Quality and Visual Quality are one- ability (r = 0.70), which also correlated between them
dimensional, as are ratings on each of the ve Sound Source scales. (r = 0.62).
Qualia ratings can be reduced to yield two measures with higher Loudness had positive correlations with Occupancy (r = 0.27)
reliability. In the model by Axelsson et al. [3], the eight adjectives and Human (tau = 0.24), and negative with Nature
(pleasant, exciting, eventful, chaotic, annoying, monotonous, (tau = 0.20), Pleasantness (r = 0.26), Visual Quality
uneventful, calm) are considered as equidistant and equally strong (tau = 0.19), and Priciness (r = 0.26). In other words: in the
semantic concepts in relation to soundscapes. They can be imag- surveyed restaurants, louder places had rougher sound and larger
ined as spokes 45 apart within a two-dimensional circumplex, variations in noise levels. They were more crowded and had more
so that scores for Pleasantness and Eventfulness are given by the sounds from people than from nature. They were rated as more
Cartesian coordinates: unpleasant, had worse visual quality, and were less expensive.
Pleasantness RRA  cos2p  NA =8
2.3.2. Prices and physical environment
Ev entfulness RRA  sin2p  NA =8 The correlations between Priciness and Loudness (r = 0.26),
and Visual Quality (tau = 0.26), indicate that expensive places
where RA is a vector of ratings {0 . . . 1} on the one-dimensional were less loud and had visually better environments. There was
scales and NA is a whole number {0 . . . 7} corresponding to the a moderately strong correlation between Priciness and Fluctuation
adjectives index, as listed above. strength (r = 0.26), and the amount of Fan sound sources corre-
Results for these scores are listed in Table 2b, together with rat- lated with F50 (tau = 0.21), but Fan amount was not associated
ings on the ve Sound Source scales, and the two scales for Sonic with Priciness, as might have been expected. This gave rise to the
and Visual Quality. question if another sound source could contribute to uctuation
strength and at the same time be (negatively) correlated with price
2.3. Correlation analysis levels. It is conceivable that vocal communication (cf. [13]) could
do this, but it must be emphasised that uctuation strength is
The relationships between psychoacoustic, physical, and per- not a validated descriptor of the relative amount of speech among
ceptual features were investigated with correlation analysis. The other sounds sources. In the survey, the correlation between Prici-
median values of psychoacoustic descriptors (N50, S50, R50, ness and Human was non-signicant, indicating that the amount of
F50) and on-site LAeq1m (henceforth LAeq) were included, human-origin sounds was not related to the price level of the res-
together with Loudness Variability (N1090), which Axelsson et al. taurants. These intriguing results could be due to some other,
[3] had found to correlate with Pleasantness. unidentied factor that we have not considered. Moreover, we

Table 2a
Statistics for acoustic and psychoacoustic descriptors in 112 restaurants.

LAeq LCeq N50 S50 R50 F50 N1090 S1090 R1090 F1090
Mean 70.2 76.1 20.8 1.27 0.0527 38 6.97 0.268 0.0707 9.52
SD 4.74 4.41 6.81 0.084 0.0097 11.6 3.42 0.0815 0.0193 3.26
Median 70.3 76.2 19.5 1.26 0.0524 34.7 6.2 0.26 0.0657 8.79
Min 57.3 66.5 8.91 0.935 0.0319 22.6 1.82 0.148 0.0402 4.17
Max 81.9 88.6 49.8 1.53 0.0911 88.9 26.4 0.885 0.169 21.1

LAeq = equivalent A-weighted SPL (dB); LCeq = idem C-weighted; N50 = median Loudness (sone); S50 = median Sharpness (acum); R50 = median Roughness (asper);
F50 = median Fluctuation strength (magnitude, see Section 2.2.2); N1090 = loudness variability, the difference between 10th and 90th percentiles of Loudness envelope;
S1090 idem of Sharpness; R1090 = idem of Roughness; F1090 = idem of Fluctuation strength.
52 P. Lindborg / Applied Acoustics 92 (2015) 4760

Table 2b
Statistics for physical features and perceptual ratings in 112 restaurants.

Priciness (bx) Occupancy (%) Pleasantness Eventfulness Sonic quality Visual quality Nature Human Trafc Fans Other noise
Mean 1.42 52 0.13 0.21 2.90 3.42 1.28 3.85 1.83 2.20 3.09
SD 0.46 96 0.66 0.63 1.01 1.09 0.63 0.90 1.22 1.17 1.30
Median 1.38 38 0.141 0.271 3 4 1 4 1 2 3
Min 0.00 2 1.79 1.51 1 1 1 2 1 1 1
Max 2.53 100 1.45 1.73 5 5 4 5 5 5 5

For Priciness and Occupancy, see Section 2.2.3. Perceptual ratings as in the Swedish Soundscape Quality Protocol [4].

surmise that the Sound Sources protocol did not lead raters to dis- Trafc correlated with Fan (tau = 0.28) and, as we have seen,
tinguish between a background din of voices and a foreground spo- Fan with Fluctuation strength (tau = 0.21). In other words, res-
ken conversation. The former might presumably contribute taurants tended to be dominated either by a mix of trafc and nat-
negatively to quality scores, and the latter positively. Further ure sounds, or by human sounds.
research is needed to investigate the relation between Priciness
and non-designed human sources, in particular vocal utterances
of different valence, such as loud talking versus laughter. 2.4. Summary and discussion of the correlation analysis
Occupancy correlated positively with Loudness (r = 0.27) and
Loudness Variability (r = 0.33). Unsurprisingly, it was associated To sum up, the correlation analysis revealed patterns of rela-
with the amounts of Human and Natural sound sources and in tions between quality ratings, descriptors, and physical features.
opposite directions (tau = 0.36 and 0.26, respectively). Thus, As for the research question Are prices and noise levels related?,
the more densely crowded places in our survey had noisier and conclusions are at this point tentative, given that the measure for
more varying ambiences, with sounds identied as coming from price levels, Priciness, was developed as part of this investigation
people rather than from natural sources. and further work is needed to validate the approach. The results
Additionally, the association between Priciness and Occupancy for the surveyed restaurants showed that less expensive places
was borderline signicant (r = 0.19, p = 0.046) in the survey, which had higher noise levels. Furthermore, Priciness was associated
might indicate that expensive places tended to be operating nearer with timbral characteristics of the ambiences, specically uctua-
maximum capacity. tion strength. This might indicate a higher prevalence of spoken
voices in less expensive places, but uctuation strength can also
2.3.3. Ratings of quality depend on sound from other sources. Moreover, as discussed in
Pleasantness was negatively correlated with Loudness Section 2.2.2, the present study used a non-standard method for
(r = 0.26) as well as with the amount of sounds from sources estimating uctuation strength, so this result must be interpreted
identied as Human (tau = 0.19), Fans (tau = 0.20), and Oth- cautiously. We believe that a measure for uctuation strength of
ernoise (tau = 0.24). Thus, pleasant places were less loud and time-varying, natural sounds (in particular complex soundscapes)
had fewer noises from people and machines. In addition, it was has yet to be developed; such a measure would be a valuable
noted that the correlation between Pleasantness and Loudness contribution.
Variability (N1090) was weakly signicant (r = 0.20, p = 0.033) Of a more general consideration, the reviewed servicescape lit-
and in the expected direction, thus lending support to results erature does not clearly state whether prices for food and beverage
reported in Axelsson et al. [3]. should be considered a factor that the restaurant manager has con-
The only signicant correlation involving Eventfulness was with trol over on par with other servicescape design elements, e.g. light-
LAeq (r = 0.23). Since the A-lter boosts levels in the 2 . . . 4 kHz ing or interior design materials (this would make it an independent
frequency region, it could be that this measure is more sensitive factor in a regression analysis) or if it is a dynamic factor that, per-
than N50 in capturing features linked to Eventfulness, such as haps, reects the popularity of a place (this would make it a
speech. However, the association was fairly weak (p = 0.015) and response variable). To investigate this topic further, an analyst
will not be pursued here. might have to work with a more elaborate theory of multimodal
As expected from earlier results [26], Sonic Quality and Visual restaurant design and branding, that would generate a testable
Quality correlated strongly (tau = 0.41), and also with Pleasant- hypothesis related to quality perception.
ness (tau = 0.46 and 0.39, respectively). A separate test As for How important is the soundscape to the experience of
revealed that the restaurants were consistently rated higher on the whole environment?, the results showed that ratings of Visual
Visual than Sonic Quality (Wilcoxon signed rank V = 327, Quality and Sonic Quality shared patterns of relations with psy-
p = 4e06), with a medium-sized effect of almost 0.5 standard choacoustic descriptors and sound sources, but that the latter
deviations. This result could indicate that sonic environments gen- had weaker correlations than the former. However, both correlated
erally are worse than visual environments in Singaporean restau- strongly with Pleasantness scores. Overall, Visual Quality was sig-
rants, or, alternatively, that the raters were more critical to nicantly higher than Sonic Quality in the surveyed restaurants. If
sounds than sights in the surveyed restaurants. this is a real effect, it reveals a need to pay more attention to sound
For both kinds of quality ratings, the pattern of associations in the design of restaurant environments: this is a business oppor-
with other variables were similar. Visual Quality correlated with tunity. Conversely, the effect could be due to rater bias. Since the
Human and Natural sound sources (tau = 0.21 and 0.25, raters were aware that soundscape quality was the focus of the
respectively) as well as with Loudness (tau = 0.19). For Sonic survey, they might have been more critically inclined in this
Quality, these correlations had the same sign, but were generally regard. The question of the relative importance between sound
weaker. and visuals (as well as other sensorial input) might be probed with
a factorial experiment design in a VR simulation of restaurants,
2.3.4. Sound sources where auditive and visual aspects of the environment can be
The amount of Natural sounds correlated with that of Human modied separately and their contributions to overall quality
sounds and Trafc (tau = 0.32 and 0.32, respectively). Also, assessed.
Table 3
Cross-correlation matrix for multimodal measurements in 112 restaurants, showing statistics (Pearsons r or Kendalls tau) with associated probability (p < 0.001,
p < 0.01, p < 0.05).

Psychoacoustic descriptors Physical features Ratings of quality Sound sources

LAeq (bx) N50 (bx) S50 R50 (bx) F50 (bx) Loudness Priciness (bx) Occupancy Pleasantness Eventfulness Sonic Visual Nature Human Trafc Fan
Variability (bx) (bx) quality quality
N50 0.82
p = 0
S50 (tau)
R50 0.56 0.71
p = 0 p = 0
F50
Loudness 0.55 0.70 0.62
variability p = 0 p = 0 p = 0
Priciness 0.26 0.26
p = 0.0062 p = 0.0055

P. Lindborg / Applied Acoustics 92 (2015) 4760

Occupancy 0.31 0.27 0.24 0.32 0.19
p = 0.00091 p = 0.0045 p = 0.011 p = 0.00049 p = 0.046
Pleasantness 0.21 0.26 0.20
p = 0.023 p = 0.0063 p = 0.035
Eventfulness 0.23 0.19 0.21
p = 0.014 p = 0.049 p = 0.024
Sonic quality 0.15 0.16 0.14 0.46
(tau)
p = 0.042 p = 0.023 p = 0.047 p = 0
Visual Quality 0.19 0.16 0.26 0.16 0.39 0.41
(tau)
p = 0.0082 p = 0.02 p = 0.00021 p = 0.025 p = 0 p = 0
Nature (tau) 0.16 0.20 0.26 0.25
p = 0.037 p = 0.009 p = 0.00087 p = 0.0026
Human (tau) 0.28 0.24 0.22 0.36 0.19 0.16 0.21 0.32
p = 8.2e-05 p = 0.001 p = 0.0018 p = 0 p = 0.0079 p = 0.046 p = 0.0098 p = 0.00017
Trafc (tau) 0.15 0.18 0.32
p = 0.041 p = 0.028 p = 0.00023
Fan (tau) 0.21 0.20 0.17 0.28
p = 0.0028 p = 0.0054 p = 0.027 p = 0.00052
Othernoise 0.17 0.24 0.17 0.16
(tau)
p = 0.016 p = 0.00055 p = 0.032 p = 0.044

For clarity, only correlations with p 6 0.05 are included. N50 = Loudness, S50 = Sharpness; R50 = Roughness; F50 = Fluctuation strength. (bx) = Box-Cox transformed variables; (tau) = Kendalls tau, used for non-parametric
correlations where normality of distributions could not be assumed.

53
54 P. Lindborg / Applied Acoustics 92 (2015) 4760

2.5. Analysis of restaurant categories Table 4

Types of restaurants in the survey, by Design Style and Food Style categories.

We now turn to the question of how ratings, descriptors, and Chinese MixFusion OtherAsian Western Sum
physical features differ between categories of places. Taking indoor Bar&Buffet 0 4 2 1 7
and outdoor Situation under one, the analysis involved two classi- Caf 6 5 3 11 25
cations, by Food Style and Design Style. FastFood 3 5 2 11 21
For Food Style, the annotation protocol offered seven a priori Hawker 5 13 4 0 22
Dining 7 4 12 14 37
dened categories (see Section 2.1). As it happened, the annotators Sum 21 31 23 37 112
frequently chose to be specic, employing in all 32 different attri-
butes for the 112 restaurants. 62% of places were labelled as either
American, Chinese, International, Japanese, or Western. The author
grouped places with annotations denoting a mixed yet exclusively and meaningful attributes in this context. The most common were:
Asian cuisine, e.g. Chinese, Malay, Indian, in a new category Oth- table (80 times), chair (73), plastic (61), wood (58), oor (25), con-
erAsian, and places with annotations denoting an East-West cross- crete (22), wall (20), glass (17), tile (15), furniture (9), and lami-
over menu, e.g. Chinese, Indian, Western or International in a nate, marble, metal, and steel (8). In this analysis, the focus was
new category MixFusion. The single places labelled as British, Can- directed to words indicating the materials used, and therefore
ton[ese], Dim Sum (Hong Kong), French, Fusion, Indonesian, Italian, words describing furniture, eating utensils, and physical dimen-
Korean, Local, Mix[ed], Thailand and Vietnam[ese] could easily be sions of the room were discarded. This left 30 unique terms, used
sorted into these broader categories. This yielded a revised classi- 255 times in total, which the analysist clustered into six types:
cation by Food Style in four categories: Chinese (21 places), Mix- Plastic (plastic, berglass), wood (wood, plywood, laminate, chip-
Fusion (31), OtherAsian (23), and Western (37). They were of board, rattan, parquet), Stone (concrete, stone, pillars, rock, brick,
similar size (X2(3) = 5.9, p = 0.12 n.s.). ceramic, tile, marble), glass (glass, glasspanel), metal (metal, steel,
For Design Style, the 112 places were originally annotated as: aluminium), and textile (canvas, cloth, curtain, carpet, sofa, para-
Bar (3 places), Buffet (4), Canteen (10), Coffee or Tea House (25) sol, cushion). The two remaining terms (sand, beach) only
FastFood (11), FoodCourt (2), Hawker (19), NightMarket (1), and appeared in the Bar&Buffet category and were excluded.
Restaurant (37). It was found reasonable to merge FoodCourt and Figs. 1a and 1b show the frequencies of annotations of materials
Hawker since these are essentially synonyms referring to a in restaurants by Food Style and by Design Style. Note that Plastic,
south-east Asian version of the multi-unit Quick Service Restau- Wood, and Stone were more common overall than Glass, Metal, or
rant [32]; however, the historical necessities that gave rise to Textiles.
the Asian food court are likely to have been different). The single
place annotated NightMarket was put in this category. Under Can-
teen appeared several fast-food places on the University Campus, 2.6.1. Regression onto categories
often part of a franchise as down-town places categorised as Fast- The relationships between frequencies of materials and the cat-
Food (for example MacDonalds); this provided a rationale for egorisations were investigated with multinomial logistic regres-
merging these two categories. Note that there is a clear difference sion using a neural net approach and the R package nnet [44].
between a foodcourt and a fast-food place. While the former con- This technique can be understood as an extension of binomial
tains several small food vendors (called hawkers) spread out logistic regression to cases with more than two levels of a categor-
around a central, common eating area with tables, the latter is ical response variable. The method compares the vector of predic-
run by a single company. Finally, Bar and Buffet were merged tors between all the pair-wise combinations of the levels of the
and Restaurants were re-baptised Dining to avoid confusion. This response. The corresponding multiple Wald tests for pair-wise
yielded a revised classication by Design Style in ve categories: dichotomous differences is similar to Tukeys Honest Signicance
Bar&Buffet (7 places), Caf (25), FastFood (21), Hawker (22), and Difference test, but for the case of a multinomial instead of a con-
Dining (37). The rst category contained a smaller sample than tinuous response variable.
the others. Excluding it, the four others were of similar size Results for the multinomial logistic regression of six materials
(X2(3) = 6.2, p = 0.10 n.s.). onto Design Style revealed no signicant differences between
See Table 4 for an overview of the number of places in each cat- Bar&Buffets, Cafs, or Dining (there was borderline more Wood
egory. Note that the contingency table does not have cells of sim- in Bar&Buffet than in Dining, with Walds z = 1.96, p = 0.050). Note
ilar size (X2(12) = 33, p = 0.000); for example, hawker stalls could that the small category of Bar&Buffets had no annotations of Glass
never be entirely Western, although they often include a Western or Metal. Likewise, there were no signicant differences between
Food stall. Therefore, separate analyses were made (cf. [11]). the Hawker and FastFood categories. Note that Hawker places
had no Glass, and FastFood places had no Textiles.
2.6. Interior design materials Hawkers had signicantly more Plastic than Bar&Buffet
(z = 2.71), Caf (z = 3.09), and Dining (z = 2.73) places. They
While the categorisation by Food Style was based on observa- had more Metal than Bar&Buffet and Dining (z = 2.21). FastFood
tions of menu items, the categorisation by Design Style was based places had more Plastic than Bar&Buffet (z = 2.03), and more
on observations of the physical environment, specically architec- Metal than Cafs (z = 2.00) or Dining places (z = 2.46).
ture and materials. One research question was Considering their To sum up, there were signicant differences between Design
acoustic impact, which interior design materials are common in Style categories in terms of interior materials. Analysis of free-form
restaurants?, and the collectors had been instructed to make text annotations revealed that the surveyed restaurants split into
free-form notes of elements which could be suspected to have an two main groups: on the one hand, Hawkers and FastFood places
inuence on room acoustics. Their annotations were analysed with had more Plastic and Metal, and on the other hand, Cafs, Dining,
a quantitative linguistic analysis approach using functions in the R and Bar&Buffets had more Textiles.
package koRpus [29]. First, irrelevant words (e.g. it, and and so Results for the multinomial logistic regression of six materials
forth) were excluded, and plurals were collapsed with singulars. onto Food Style showed that the amount of Stone was signicantly
Then, alternative spellings were corrected manually. This yielded higher in places with Chinese style than in places where Other-
a list of 592 single-word annotations, out of which 78 were unique Asian or Western food was served (z = 2.06 and 2.16, respec-
 P. Lindborg / Applied Acoustics 92 (2015) 4760 55

Interior Design Materials by FoodStyle (95% conf.int.) side the training set is taken as input, producing a response by
maximum likelihood estimation. If this estimate corresponds with
Plastic the target (i.e. the known, true value), the prediction is considered
1.4

Wood
Stone successful. For example, after training the model on a subset of
Glass the restaurant data, a vector representing the number of annota-
1.2

Metal
Textile
tions of each material (Plastic, Wood, Stone, Glass, Metal, Textile)
at one particular location yielded its most likely Food Style or
word counts per location

Design Style, which would be correct or not. In the analysis, a ran-

1.0

dom sample of 101 restaurants (approximately 90% of all) was

used to train the neural net model, and then to predict Style in
0.8

the remaining 11 restaurants. Sub-sampling was repeated

5000 times for each model, and mean prediction success rates
0.6

were formed by averaging.

This method applied to Food Style achieved an overall success
rate of 34%, higher than chance (which would be 25%). Success
0.4

rates were simultaneously calculated for each category separately,

with correction for different category size. The model was best at
0.2

predicting MixFusion places (58% of these), then Western (47%),

and less successful with Chinese (9%) and OtherAsian places (5%).
It was found that the pattern of prediction success changed when
0.0

Chinese MixFusion OtherAsian Western sets of smaller numbers of predictors were tested. An exhaustive
FoodStyle analysis of all 63 combinations of the 6 predictor variables revealed
that many combinations yielded high average prediction rates for
Fig. 1a. Mean frequencies, with 95% condence intervals, of six types of interior MixFusion and Western places, and much lower for Chinese and
design materials (see legend) in 112 restaurants, by categories of Food Style.
OtherAsian places. This might indicate that the two former catego-
ries were more specic in terms of the materials they featured,
while the two latter categories featured materials in more random
Interior Design Materials by DesignStyle (95% conf.int.)
ways. The latter categories might be better characterised by some
Plastic other materials that the present analysis did not include.
1.5

Wood Similarly, applying the prediction method to Design Style

Stone
Glass achieved an overall success rate of 32%, higher than chance
Metal (20%). The simultaneous size-corrected success rates for separate
Textile
categories were highest for Dining (47%), FastFood (42%), and Haw-
word counts per location

ker (34%); and much lower for Cafe (6%) and Bar&Buffet (0%). The
rst three might thus be said to have more specic interior design
1.0

materials than the two latter. However it must be recalled that the
number of places in the Bar&Buffet category was signicantly
smaller than in the others. A larger sample should allow better pre-
dictions for such places.
To sum up, the categories could be predicted overall from inte-
0.5

rior design materials alone at higher than chance level. The classi-
cations by Food Style and Design Style, that had been made on-
site and simplied in a previous part of the analysis, were thus
supported.

2.7. Analysis of variance between categories

0.0

Bar&Buffet Cafe Fastfood Hawker Dining

To analyse the how psychoacoustic descriptors, physical fea-
DesignStyle
tures, and perceptual ratings vary among the categories, we per-
Fig. 1b. Mean frequencies, with 95% condence intervals, of six types of interior formed multiple analyses of variance (MANOVA), with Food Style
design materials (see legend) in 112 restaurants, by categories of Design Style. and Design Style as independent, categorical variables. This inves-
tigation had been planned a priori and the signicance level for
tively). Western food places had borderline less Plastic than MixFu- each test was set at alpha2-tailed = 0.05. Results are given in Table 5.
sion places (z = 1.93, p = 0.054). Note that OtherAsian places had no Categorisations with signicant differences were further explored
Glass. with Tukeys Honest Signicant Difference test for multiple com-
parisons. The results are discussed below and illustrated in the
2.6.2. Predicting categories from materials plots in Figs. 2a and 2b.
To further the investigation of the usage of materials in interior
design, we analysed if restaurant categories could be predicted 2.7.1. Results for the planned comparisons
directly from the annotations. We followed a method for multino- 2.7.1.1. Psychoacoustic descriptors. In both Food Style and Design
mial prediction described in the IDRE webpages [22], and for cross- Style categories, there were signicant differences for LAeq and
validation applied a repeated random sub-sampling approach in N50. As seen earlier, these two measures correlated strongly over-
order to achieve a more robust assessment of how generalisable all (r = 0.82). Tukeys HSD test revealed that places in the Food
such a prediction would be. In cross-validation, rst, a subset of Style category Western (37 out of the 112 places) were signi-
the data is used to train a model (in our case the same neural cantly less loud than the others, with a difference of 3.4 sone (for
net as described in Section 2.1.1, nnet; [44]. Then, a data point out- LAeq, the difference was 2.7 dB).
56 P. Lindborg / Applied Acoustics 92 (2015) 4760

Table 5
Multiple analysis of variance (MANOVA) in restaurants categorised by Food Style and Design Style. For variable names, see Table 2 and Sections 2.1 and 2.2. Effect size is Cohens d
with pooled standard deviation estimate. (p < 0.001, p < 0.01, p < 0.05).

Food Style Design Style

F p Effect size F p Effect size
LAeq 2.99 0.034 0.08 2.16 0.079.
N50 1.86 0.14 2.8 0.030 0.10
S50 0.222 0.88 1.31 0.27
R50 2.16 0.098. 1.79 0.14
F50 1.82 0.15 1.26 0.29
N1090 0.952 0.42 0.317 0.87
Priciness (bx) 4.63 0.0044 0.12 19.4 0 0.43
Occupancy (bx) 0.903 0.44 2.22 0.072.
Pleasantness 0.121 0.95 1.61 0.18
Eventfulness 0.619 0.60 2.21 0.073.
Nature 0.867 0.46 4.24 0.0032 0.14
Human 0.627 0.60 3.25 0.015 0.11
Trafc 1.55 0.21 4.12 0.0039 0.14
Fan 6.02 0.0008 0.15 1.52 0.20
Othernoise 0.303 0.82 0.187 0.94
Sonic Quality 0.205 0.89 3.12 0.018 0.11
Visual Quality 0.0258 0.99 5.61 0.0004 0.18

2.7.1.2. Prices and physical environment. Unsurprisingly, the small cantly more expensive than Chinese (d = 0.40 SD). Occupancy did
group of Bar&Buffet places was the most expensive. The HSD tests not differ signicantly between categories.
revealed that it was signicantly higher in Priciness than Dining
(difference = 0.45 standard deviations), which in its turn was sig- 2.7.1.3. Ratings of quality. Neither Pleasantness nor Eventfulness
nicantly higher than the remaining three categories (d = 0.46 was signicantly different between categories. In terms of Design
SD). Moreover, the test showed that Western places were signi- Style, the patterns for ratings of Sonic Quality and Visual Quality

LAeq N50 Priciness

2.5

74 25
2.0
72
20
70
1.5
68
15
66 1.0

Chin MixF Asia West Chin MixF Asia West Chin MixF Asia West

Nature Human Traffic

5 5 5

4 4 4

3 3 3

2 2 2

1 1 1

Chin MixF Asia West Chin MixF Asia West Chin MixF Asia West

Fan Sonic Qual Visual Qual

5 5 5

4 4 4

3 3 3

2 2 2

1 1 1

Chin MixF Asia West Chin MixF Asia West Chin MixF Asia West

Fig. 2a. Means and 95% condence intervals for selected features of 112 restaurants, by categories of Food Style. Chin = Chinese; MixF = MixFusion; Asia = OtherAsian;
West = Western; LAeq = on-site A-weighted SPL (dB); N50 = median Loudness (sone). For other variables, see Sections 2.1 and 2.2.
 P. Lindborg / Applied Acoustics 92 (2015) 4760 57

LAeq N50 Priciness

2.5

74 25
2.0
72
20
70
1.5
68
15
66 1.0

BB FF Hawk Din BB FF Hawk Din BB FF Hawk Din

Nature Human Traffic

5 5 5

4 4 4

3 3 3

2 2 2

1 1 1

BB FF Hawk Din BB FF Hawk Din BB FF Hawk Din

Fan Sonic Qual Visual Qual

5 5 5

4 4 4

3 3 3

2 2 2

1 1 1

BB FF Hawk Din BB FF Hawk Din BB FF Hawk Din

Fig. 2b. Means and 95% condence intervals for selected features of 112 restaurants, by categories of Design Style. BB = Bar&Buffet; Caf = Caf, FF = FastFood; Hawk = Hawker;
Din = Dining; LAeq = on-site A-weighted SPL (dB); N50 = median Loudness (sone). For other variables, see Sections 2.1 and 2.2.

were similar (cf. Section 2.3.3), and there were signicant differ- ference was 0.64 scale steps on the 5-degree Likert scale of the pro-
ences between Design categories. This is further discussed in tocol (d = 0.62), and for Sonic Quality, it was 0.61 steps (d = 0.58
Section 2.7.2. SD). For the amount of Trafc sounds, the contrast was 0.86 steps
(d = 0.76 SD). The effect sizes in all cases above can be considered
2.7.1.4. Sound sources. In terms of Food Style, the MixFusion cate- as medium size.
gory (31 places) stood out for having signicantly more Fan The colour plot in Fig. 3 visualises the 112 surveyed restaurants,
sounds. In terms of Design Style, Tukeys test showed that Bar&Buf- categorised by Food and Design Styles. The x-axis represents
fet places had signicantly more sounds of Nature origin and fewer Sound Level (LAeq was chosen over N50 since it is more familiar),
sounds heard as Human. the y-axis represents Quality (calculated as the average of z-
scaled Pleasantness, Sonic Quality, and Visual Quality ratings),
2.7.2. Post-hoc contrasts and symbol sizes are proportional to Priciness. Note that larger
Inspection of the barplots revealed two patterns of difference symbols tend to be found in the upper left quadrant, i.e. the more
between categories that led to making post-hoc contrast compari- expensive restaurants in the survey were less loud and more
sons. Effect size of the difference between group means was calcu- pleasant.
lated with Cohens d, with pooled standard deviation estimate. In a nal analysis of contrasts, we combined the two groupings
Fig. 2a suggested that Western places (37 restaurants) were less that were identied for Food Style and Design Style, respectively,
loud than Chinese, MixFusion, and OtherAsian grouped together as discussed above. One group thus consists of Western Bar&Buffet,
(75 places in all). The difference for N50 was 3.4 sone (Cohens FastFood, and Dining places (25 restaurants), and the other of Chi-
d = 0.51 SD), and for LAeq it was 2.7 dB (d = 0.59 SD). nese, MixFusion, and OtherAsian places designed in the styles Caf-
Fig. 2b suggested that on the one hand, Bar&Buffet, FastFood, s and Hawkers (36 places). The differences between means of these
and Dining (58 places) could be grouped together, and on the contrast groups were: for LAeq = 4.4 dB (Cohens d = 1.02 SD), for
other, Cafs and Hawkers (47 places). An M-like pattern was evi- Quality = 0.47 (d = 0.65 SD), and for Priciness = 0.85 (d = 1.37 SD).
dent for N50, LAeq, Trafc, and Fan, as well as a W-like pattern The effect sizes can be considered large for LAeq and Priciness,
for Sonic Quality and Visual Quality. For N50, the contrast between and medium for Quality.
these two groupings was 4.1 sone, with an effect size of 0.60 SD, Two ellipses have been drawn in Fig. 3 in order to highlight
and for LAeq, it was 2.4 dB (d = 0.55 SD). For Visual Quality, the dif- these differences. The radii of the black ellipse represent 95% con-
58 P. Lindborg / Applied Acoustics 92 (2015) 4760

Fig. 3. Plot of 112 Singaporean restaurants by selected features. X-axis = Sound Level (LAeq); Y-axis = Quality (mean of z-scaled Pleasantness, Sonic Quality, and Visual Quality);
Symbol size = Priciness. Symbol shapes code Design Styles and colours code Food Styles; see legend. The two ellipses have radii representing the 95% condence intervals
around the means of the corresponding measures, respectively, for two contrast groups dened post-hoc: Western Bar&Buffet, FastFood, and Dining places (black, upper left);
and Chinese, MixFusion, and OtherAsian places in Design Style Caf and Hawker (purple, lower right).

dence intervals around the means of LAeq and Quality for the rst might inform predictive models of environmental quality in
group, and similarly, the purple ellipse represents condence inter- restaurants.
vals around the means for the second. As for What differences are there between various types of res-
taurants?, the analysis revealed signicant differences between
restaurant categories in terms of Loudness, Priciness, Visual and
Sonic Quality, and Sound Sources. Post-hoc comparisons showed
2.8. Summary and discussion of the analysis of categories that Bar&Buffet, FastFood, and Dining formed one group, and Cafs
and Hawkers another. The former group tended to be less loud and
Summing up, the categorisation by Design Style and Food Style more expensive, and to have higher quality ratings as well as fewer
were slightly modied to better capture the on-site observations. sounds from trafc and fans.
To answer the research question Which interior design materials There was a greater number of features that distinguished
are common in restaurants?, a lexical analysis of annotations among Design Styles (both acoustic, physical, and perceptual) than
was performed. It revealed that Hawkers and FastFood places among Food Styles (no acoustic differences) in the survey. This
had signicantly more Plastic and Metal, and that Bar&Buffets, could imply that establishing a restaurants identity through
Cafs, and Dining restaurants had signicantly more Textiles. This design is stronger than through food. Note however that Priciness
is in line with Frids observations (2013). While Stone was a rela- differed between categories in both classications.
tively common material in places with Chinese food on the menu,
Plastic was less common in Western places.
Several authors have remarked that contemporary restaurant
interiors are dominated by hard surfaces and open space architec- 3. Conclusions
ture (e.g. [40,18]), with obvious implications for room acoustics, in
particular reverberation time and vocal communication comfort This article has presented results from a survey of 112 Singap-
[38]. Further empirical research into the relative proportions of orean restaurants focussing on the sonic environment. We have
interior design materials, taking categorisations into account, employed established methods for psychoacoustic descriptors
 P. Lindborg / Applied Acoustics 92 (2015) 4760 59

(such as Loudness; [10]), quality assessment (such as Pleasantness; be implemented on a handheld device and integrate audiovisual
[3], and physical factors (such as Occupancy; [24]. data capture, on-board calculation of psychoacoustic descriptors,
and ratings of environmental quality, food, and service: a merge
3.1. Contributions of NoiseTube [27], [35]) and buUuk [7].
Our survey of restaurants in Singapore has revealed that less
We have developed a novel method to determine relative Prici- loud restaurants had higher prices, and that higher prices were
ness in a sample of restaurants. Contributing to theory building, we not associated with lower occupancy (if anything, rather the oppo-
have advanced a typology of designed and non-designed sonic ele- site). Moreover, sonic quality in the surveyed restaurants was sig-
ments that might be useful for analysing and describing restaurant nicantly much worse than visual quality. For managers and
soundscapes. acoustic designers, these ndings point to an opportunity. Firstly,
Results from the correlation analysis revealed signicant associ- there is more room for improvement of the soundscape in restau-
ations between psychoacoustic, physical, and perceptual features rants than of the corresponding visual environment. Secondly, a
in the surveyed restaurants. These have been discussed in Sec- reduction in ambient noise levels is likely to lead to higher envi-
tion 2.3. The main results were that: ronmental quality ratings, both visual and sonic, and might at
the same time be a justication for an increase in menu prices.
 The louder restaurants in the survey had rougher sound and lar- Thirdly, these changes can be made while maintaining, or even
ger variations in noise levels. They were more crowded, had augmenting, the level of patronage. This three-point action plan
more sounds from people than from nature, and were rated as leads not only to more customer satisfaction, but also to larger
more unpleasant. gross margin: two incentives for managers to make the necessary
 The more expensive restaurants were less loud and had visually investments.
better environments. The analysis we have presented was based on results from Sin-
 The more pleasant restaurants had fewer sounds identied with gapore, which has climatic, cultural, economic, and other specic
people or machine sources. characteristics. The conclusions are thereby limited in generalis-
ability, and whether they can be extended to other locale remains
These results indicate that sounds of nature might abate the to be investigated.
negative inuence of mechanical noises in restaurants, thus lend- In ne, we would like to suggest that a considered selection of
ing support to the hypothesis that sound source identication interior design materials for restaurant construction and renova-
plays a crucial part in the perceived valence of soundscapes (cf. tion can, in addition to improving the acoustic and visual environ-
[34,2]). Sounds imitating nature, such as recorded bird song or ment, support the overall perception of the environment in terms
water fountains (i.e. biophony simulacra and geophony simula- of Design Style and Food Style, thus strengthening the brand identity
cra; see Table 1) are often present in servicescapes (cf. [20,21]). of the establishment as a whole.
It is an open question how effective they are in moderating per-
ceived pleasantness in comparison with actual, real sources, or Acknowledgements
for that matter, with other sonic design elements. Humans repre-
sent another complex source of sounds in restaurants; while a The author would like to acknowledge the contribution to data
background din of voices might lead to negative appraisal, the collection made by his students: T Boey, J Chai, E Cheong, H Choo, J
occasional laughter and a sustained vocal communication comfort Chua, E Im, P Kho, V Koh, S Koh, C Leong, K Lim, Y Liu, N Tan, TT
would be positive. Nguyen, A Nur, M Ong, Y Santhiramohan, D Toh, and TP Voh.
One distinction made in the typology (see Table 1) concerns
interior design materials impacting on acoustics. In the survey, this
aspect was investigated through a linguistic analysis of free-form References
architecture & materials annotations made on site. A predictive [1] Anderson J Restaurants: Dining on Decibels; 2008. <http://ergoweb.com/
model gave support to the classication in categories of Design restaurants-dining-on-decibels/> (accessed 07.12.13).
Style and Food Style. Analysis of variance revealed signicant dif- [2] Asutay E, Vstfjll D, Tajadura-Jimnez A, Genell A, Bergman P, Kleiner M.
Emoacoustics: a study of the psychoacoustical and psychological dimensions
ferences between categories in terms of loudness, quality, and
of emotional sound design. J Audio Eng Soc 2012;60(1/2):218.
price levels for both types of classications. [3] Axelsson , Nilsson ME, Berglund G. A principal components model of
A post-hoc contrast analysis suggested a simplied classica- soundscape perception. J Acoust Soc Am. 2010;128(5). 2836-2346.
tion, with Bars & Buffets, FastFood, and Dining places comprising [4] Axelsson , Nilsson ME, Berglund G. Swedish Soundscape Quality Protocol (v.
Mar. 2011). Personal communication with Axelsson; 2011.
one group, and Cafs and Hawker stalls another. More research is [5] Berglund B, Nilsson ME, Axelsson , Soundscape psychophysics in place. In:
needed to investigate this scheme further. It could be that restau- Proceedings of InterNoise; 2007. p. 10714.
rants of the rst kind are more elaborate in terms of interior design [6] Bitner MJ. Servicescapes: the impact of physical surroundings on customers
and employees. J Market 1992;56(2):5771.
in general, including acoustic elements, which leads to loudness [7] Buuuk, buUuk restaurant and bar guide; 2014.<https://itunes.apple.com/sg/
attenuation and more pleasant environments. While many Wes- app/buuuk-restaurant-bar-guide/id297106176?mt=8> (accessed 01.06.14).
tern-style fast-food franchises (e.g. McDonalds) have ceiling panels [8] Cabrera D, Psysound3; 2014 <http://www.densilcabrera.com/wordpress/
psysound3/> (accessed 01.06.14).
that are acoustically absorbent, Asian-styled hawkers and cafs in [9] Cabrera D, Ferguson S, Schubert E. PsySound3: an integrated environment for
Singapore, which might in other aspects be comparable, often the analysis of sound recordings. In: Acoustics 2008: Conference Proceedings
employ very basic interior designs that are rarely favourable for of the Australian Acoustical Society, Geelong, Australia; 2008.
[10] Chalupper J, Fastl H. Dynamic loudness model (DLM) for normal and
acoustics. Thus, the psychoacoustic and perceptual differences that hearingimpaired listeners. Acta Acust united Ac 2002;88:37886.
the present survey has uncovered might be especially pronounced [11] Christie LH, Bell-Booth RH, Acoustics in the hospitality industry: a subjective
in Singapore. and objective analysis; 2006. <http://www.victoria.ac.nz/cbpr/publications/
documents/acoustics-in-the-hospitality-industry.pdf> (accessed 07.12.13).
[12] Dag O, Asar O, Ilk O. AID: An R Package to Estimate Box-Cox Power
3.2. Implications and limitations Transformation Parameter (v. 1.1); 2012, <http://cran.r-project.org/web/
packages/AID/> (accessed 01.11.13).
Researchers might consider developing the protocol for anno- [13] Daniel P. Psychoacoustical roughness. Ch. 19 in Handbook of Signal Processing
in Acoustics, Springer New York; 2009, p. 263274.
tating sound sources used in this study into a more targeted tool [14] Daniel P, Weber R. Psychoacoustical roughness: implementation of an
for empirical studies of restaurant soundscapes. Such a tool could optimized model. Acoustica 1997;83:11323.
60 P. Lindborg / Applied Acoustics 92 (2015) 4760

[15] Ertzberger BK, The Effect Of Ambient Factors In The Design And Branding Of [30] MoM [Ministry of Manpower, Singapore]. Work in Noisy Environment ProBE;
The Restaurant Environment: A Business Approach Model Exemplied In Caf 2007. <http://www.mom.gov.sg/workplace-safety-health/resources/factsheets-
Cabernet, Tallahassee, Florid. Electronic Theses, Treatises & Dissertations. circulars/Pages/work-in-noisy-environment-probe.aspx> (accessed 11.12.13).
Florida State University; 2009. [31] MoM [Ministry of Manpower, Singapore]. Workplace Safety and Health
[16] European Parliament. Directive 2003/10/EC on the minimum health and safety (Noise) Regulations. Ch. 354A, Workplace Safety & Health Act.; 2011.
requirements regarding the exposure of workers to the risks arising from <http://www.mom.gov.sg/Documents/safety-health/WSH%20Noise%
physical agents (noise), 2003 <http://eur-lex.europa.eu/smartapi/cgi/sga_ 20Regulations%202011.pdf> (accessed 11.12.13).
doc?smartapi!celexplus!prod!DocNumber&lg = en&type_doc = Directive& [32] Muller C, Woods R. An expanded restaurant typology. Cornell Hotel &
an_doc = 2003&nu_doc = 10> (accessed. 11.12.13). Restaurant Admin Quart 1994;35(3):2737.
[17] Fastl H, Zwicker E. Psychoacoustics: facts and models. 3rd [33] Musinguzi D. The impact of restaurant noise on customers dinning experience
ed. Germany: Springer-Verlag; 2007. in Kowloon, Hong Kong. Int J Hospitality Tour Syst 2010;3(1):805.
[18] Frid E. Restaurant Soundscapes in Stockholm. Project Report TMH, KTH Royal [34] Nilsson ME. Soundscape quality in urban open spaces. In: Proc. Inter-Noise,
Institute of Technology, Stockholm; 2013. Istanbul, Turkey. Inst. Noise Control Engineering, Poughkeepsie, New York,
[19] Hall D, Irwin A, Edmonson-Jones M, Philips S, Poxon J. An exploratory USA; 2007. p. 17.
evaluation of perceptual, psychoacoustic and acoustical properties of urban [35] NoiseTube. NoiseTube Turn your mobile phone into an environmental sensor
soundscapes. Appl Acoust 2013;74:24854. and participate in the monitoring of noise pollution; 2014 <http://noisetube.
[20] Hellstrm B. Noise Design. Architectural Modelling and the Aesthetics of net/> (accessed 01.06.14).
Urban Acoustic Sapce. Bo Ejeby Frlag, Sweden; 2003. p. 1263. [36] Novak CC, La Lopa J, Novak RE. Effects of sound pressure levels and sensitivity
[21] Hellstrm B, Dyrssen C, Hultqvist A, Mossenmark S, Sjsten P. Modelling the to noise on mood and behavioral intent in a controlled ne dining restaurant
shopping soundscape. J Sonic Stud 2011:1. <http://journal.sonicstudies.org/ environment. J Culinary Sci Technol 2010;8:191218.
vol01/nr01/a04> (accessed 24.06.13). [37] OSHA [Occupational Safety & Health Administration]. Standard 1910.95,
[22] IDRE. R Data Analysis Examples: Multinomial Logistic Regression. UCLA: Occupational noise exposure; 2008. <https://www.osha.gov/pls/oshaweb/
Statistical Consulting Group; 2014, <http://www.ats.ucla.edu/stat/r/dae/ owadisp.show_document?p_table=STANDARDS&p_id=9735> (accessed
mlogit.htm> (accessed 01.10.14). 11.12.13).
[23] Lartillot O. MIRtoolbox v1.3.2; 2011, <http://www.mathworks.com/ [38] Rindel JH. Verbal communication and noise in eating establishments. Appl
matlabcentral/leexchange/24583-mirtoolbox> (accessed 24.06.13). Acoust 2010;71(12):115661.
[24] Lebo CP, Smith MFW, Mosher ER, et al. Restaurant noise, hearing loss, and [39] Rohrmann B, Soundscapes in restaurants. In: Proc. Int. Symposium Acoustic
hearing aids. Western J Med 1994;161:4550. <http://www.ncbi.nlm.nih.gov/ Ecology. Melbourne; 2003.
pmc/articles/PMC1011367/pdf/westjmed00059-0047.pdf> (accessed [40] Rusnock CF, Bush PM. Case study: an evaluation of restaurant noise levels and
07.12.13). contributing factors. J Occup Environ Hygiene 2012;9(6):10813.
[25] Lindborg PM, Physiological Measures Regress onto Acoustic and Perceptual [41] Rychtrikov M, Vermeir G. Soundscape categorization on the basis of
Features of Soundscapes. In: Proceedings of the 3rd International Conference objective acoustical parameters. Appl Acoust 2013;74:2407.
on Music Emotion (ICME3), Jyvskyl, Finland; 2013. p. 111. [42] Schafer RM. The soundscape: our sonic environment and the tuning of the
[26] Lindborg PM. (submitted). Perception of soundscapes correlates with acoustic world. 2nd ed. USA: Destiny Books; 1994. p. 1301.
features and is moderated by personality traits. [43] Truax B. Acoustic communication. USA: Greenwood Publishing Group; 2001.
[27] Maisonneuve N, Stevens M, Ochab B. Participatory noise pollution monitoring p. 1284.
using mobile phones. Inform Polity 2010;15(12):5171. [44] Venebles WN, Ripley BD. Feed-forward Neural Networks and Multinomial Log-
[28] McLaughlin K. Pass the Salt ... and a Megaphone. Wall Street Journal, 3 Feb. Linear Models, v7.3-7; 2013 <http://cran.r-project.org/web/packages/nnet/>
2010; 2010. <http://online.wsj.com/news/articles/SB10001424052748704022 (accessed. 01.03.14).
804575041060813407740> (accessed 10.12.13). [45] World Soundscape Project. Handbook for Acoustic Ecology. 2nd ed. Barry
[29] Michalke ME, koRpus, An R Package for Text Analysis, v0.04-9; 2011. <http:// Truax, editor; 1999. <http://www.sfu.ca/sonic-studio/handbook/ (accessed
CRAN.R-project.org/web/packages/koRpus/index.html> (accessed 01.02.14). 11.12.13).