Content uploaded by David Benyon
Author content
All content in this area was uploaded by David Benyon
Content may be subject to copyright.
Proceedings of the 12th International Conference on Auditory Display, London, UK, June 20-23, 2006
SOUND AND SOUNDSCAPE CLASSIFICATION: ESTABLISHING KEY
AUDITORY DIMENSIONS AND THEIR RELATIVE IMPORTANCE
Iain McGregor, Grégory Leplâtre, Alison Crerar and David Benyon
School of Computing,
Napier University,
Edinburgh, UK.
{i.mcgregor, g.leplatre, a.crerar, d.benyon}@napier.ac.uk
ABSTRACT
This paper investigates soundscape classification by using two
different forms of data gathering and two different populations.
The first method involves a questionnaire completed by 75
audio professionals. The second uses a speak-aloud
experiment, during which 40 end users were asked to describe
their audio environment. While both approaches are different
and target a different audience, they provide an indication of
key dimensions for the perception of soundscapes and their
relative importance. Contrasts and similarities between the
results of the questionnaire and speak-alouds are highlighted.
Their implications with regards to the establishment of a set of
common terms in order to aid future auditory designs are also
discussed.
1. INTRODUCTION
This paper reports upon two studies which try to establish
dimensions for a future classification of inhabited soundscapes.
This would aid the future design and evaluation of shared
immersive auditory environments through the use of a shared
language. The traditional differentiation between auditory
professionals and end users, where the former is concerned
with the quantitative soundfield, while the latter is only
concerned with the qualitative soundscape is an artificial
divide. All of us inhabit and contribute to soundscapes, and as
such, nothing can ever be designed in isolation. What can be
done, is to develop a method of better communicating the
experience.
Delage [1] points out that if end users have any chance of
interpreting the meaning of new sounds then it has “to be in the
range of what they already know”. Utilizing a classification
system based upon end user descriptions of sound events
within a soundscape can provide that insight, and also establish
where there is a mismatch between the intended design and its
final perception.
The first study was a questionnaire targeted at audio
professionals. This was originally to survey the knowledge
and practice of a wide and heterogeneous community. The
answers from 75 respondents were grouped into three
categories (designers, acousticians and computer scientists) in
order to provide a clear picture of a population that overlaps
with the ICAD community. However, this paper is only
concerned with a subset of the study that deals with what the
respondents considered to be the most relevant dimensions of
auditory environments. An attempt was made at establishing a
consensus, which was subsequently compared to end users’
experiences.
The second study involved 40 listeners, who were asked to
describe verbally what they could hear while listening to an
enclosed environment under four different conditions, with 10
participants per condition. Recordings were made of the
responses, which were subsequently transcribed and coded.
This revealed more about the relative importance of auditory
dimensions to end users in an everyday listening context. The
prioritization of these combined dimensions and their instances
into a form of classification could be used to inform the design
and evaluation of effective and meaningful sounds and
soundscapes.
2. SOUNDSCAPE CLASSIFICATION
Soundscape classification comes in a variety of forms, the
most common are based around: speech and non-speech; or
speech, music and other. A number of methods have been
developed in order to classify sound events within soundscapes
or even complete auditory environments, these can be split into
psychoacoustics, semantics, aesthetics and environmental.
2.1. Psychoacoustics
Gaver advocated an ecological approach to classifying sounds
according to their “audible source attributes”. Sound events
are generated by either solids, gasses or liquids and complex
sounds can be described by either “temporal patterning,
compound or hybrid sources” [2]. The results may be
reproduced in map form in order to illustrate the qualitative
nature of the sound events, which were heard.
Gaver acknowledged that his classification was
incomplete, citing the voice, electricity and fire as possible
additional candidates of simple sonic events. He went on to
say that any definitive classification of a source being
“somewhat questionable” due the qualitative nature of
listening. The alignment of the physical actions which
generated sound events with everyday language did provide a
form of eliciting psychoacoustical responses. There was a high
degree of potential granularity when patterned, compound and
hybrid events were included.
2.2. Semantics
In 1998 Bernard Delage collaborated with Heleen Engelen to
arrange a “Sound Design Day”, this was by invitation only,
and involved architects, acousticians, computer scientists,
composers, electroacousticians, scenographers, sound and
visual designers all of whom had sound design experience [1].
Whilst debating the role of sound and ergonomy, specifically
within the realm of auditory feedback provided by manual
tools, the group developed a list for the interactive function of
sounds. Examples included warning in terms of being careful,
assisting with regards to memory and incitement in terms of
readiness for use.
ICAD06 - 105
Proceedings of the 12th International Conference on Auditory Display, London, UK, June 20-23, 2006
Macaulay and Crerar [3] were frustrated by the lack of
appropriate auditory models for the interaction designer. In the
belief that sound reveals information by situating individual’s
inside their soundscape, rather than light which presents
information in front, they studied the work of Brewster [4],
Feld [5], Gaver [2] and Truax [6] as a basis for formulating a
soundscape classification more appropriate to the field of
Human Computer Interaction (HCI). The resultant model
provided interactive systems designers with a framework for
classifying sounds, which was a preliminary step in the move
away from contemporary visually saturated interfaces.
Macaulay and Crerar proposed a method of classifying
constituents of soundscapes based upon (i) sound type, (ii)
information category and (iii) acoustical information. The
sound type was broken down into music, speech, abstract and
everyday. (Subsequently we have found that the ‘abstract’ and
‘everyday’ concepts are more readily described as ‘other
known’ and ‘other unknown’. Moore [7] points out the
Boolean nature of the perception of sound, as either being
perceived as speech or not).
The information categories were: visible, hidden,
imagined, patterns of events, passing of time, emotions and
position in Euclidean space, which allowed an insight into the
information content provided to the soundscape inhabitant.
Finally the model included acoustical information,
(subsequently found to be the level of listening) which could
be either foreground, contextual or background. Foreground
sounds were those with which the listener actively engaged,
contextual sounds provided an underpinning to the foreground,
and background were all of the other ‘ambient’ sounds, often
not attended to [8].
2.3. Aesthetics
Gabrielsson and Sjorgen [9] set out to establish psychophysical
relationships between physical parameters such as frequency
response and perceived sound quality. They argued that
perceived sound quality should be able to be described through
“separate perceptual dimensions”. These dimensions were
ideal as a starting point for the aesthetic evaluation of sound
events as well as of sound reproduction systems.
About two hundred adjectives were given to forty sound
engineers, thirty audiologists and one hundred and five people
with hearing loss, each adjective was rated for appropriateness
when describing the perceived sound quality of speakers and
headphones in the case of the sound engineers, and hearing
aids in the case of audiologists and people with hearing loss.
This resulted in a list of around sixty adjectives being
considered suitable. The next stage was to experiment with
“normal hearing subjects” on adjective and similarity ratings as
well as free descriptions where participants were asked to use
their own vocabulary in order to describe the perceived sound
quality of a variety of sound reproduction equipment. From
this they found that there were predominately two to five
dimensions which resulted from each of the experiments, with
a final total of eight. The resultant dimensions were associated
with clarity, emotional response, spatial cues, dynamics and
spectrum.
2.4. Environmental
Amphoux [10] was concerned with the interaction of the
listener with the soundscape and developed an EMP model. E
stood for environmental listening, M for milieu listening and P
landscape listening. He argued that it is essential to consider
all three forms of listening, each of which had three categories,
with three criteria and three dimensions. Spatio-temporal was
the first of the three categories and was broken down into three
criteria: scale, orientation and atemporality. Scale represented
the comparison between physical space and perceived auditory
space with orientation referring to the ability to follow a
specific sound within the environment in three dimensions.
The fourth dimension of time, being represented through
atemporality. Semantic-cultural incorporated publicity,
collective memory and naturality. Publicity referred to the
overall impression or ‘voice’ that was presented, that of
anonymity or congregation, whereas collective memory
reflected local anomalies, which were site specific, regulating
time or even suggesting a by-gone age. Naturality concerned
the weighting of natural to ‘man-made’ sounds and whether
there were any narrative elements.
Finally sonic material referred to reverberation, sonic
signature and metabolic structure. The reverberation
incorporated the live-ness, intelligibility of reflections and the
complexity of the reflections or echoes. The sonic identity was
concerned with whether an area had unique sound such as an
unusual bell, or it was stereotypical, and if it was unusual
could it have represented a broader area such as a city or
country in the manner of a postcard. The metabolic structure
incorporated the grouping of sound sources, their relative
clarity, as well as their complexity.
Hellstrom [11] was also concerned with the concept of
place in order to study the individual identities of city quarter
soundscapes. He first broke these down into space and
character, applying traditional elements of form to complex
structures (sound groups): path, node, landmark, edge and
district and his own classification of centre, distance,
direction, tempo and rhythm for simple structures (individual
sounds). Character was split into dynamic and static, static
refers to a continuous sound from which no individual sources
could be identified, such as an interval conversation at a
concert hall where no individual conversation could be heard.
Whereas dynamic denoted a widely changing soundscape
where sounds rise and fall and are intelligible. Complex
structures could be further classified through musical
terminology: tonal-atonal, consonant-dissonant,
homogeneous-transparent, strong-weak and rhythmic-
arhythmic, with simple structures having pitch, timbre,
articulation, dynamism and duration applied. Hellstrom went
on to apply a further set of five categories to each of the
sounds identified within the overall soundscape. Each of the
categories worked in terms of opposition: Man-made vs.
natural, present vs. past, local vs. general, figure vs. ground,
and order vs. chaos.
Hellstrom applied this method to recordings he made
around Klara, in Stockholm, assessing a fountain he found in
terms of the character it possessed: a static sonic structure, and
was harmonious with hard articulation and strong intensity. In
terms of space it had space defining boundaries as well as
being a local landmark. Finally he applied three of the five
possible categories with the result that there were large
variations over day and night and was found to be a ground
object with exhibited order. Hellstrom hoped that this form of
mediation would contribute to the future design of sonic
environments.
Under the banner of the World Soundscape Project,
Schafer developed a simple terminology for describing sounds
within a soundscape as well as terms for describing the clarity
[12]. Keynote came directly from music, and was applied to
sounds that were fundamental to an environment, like traffic
on a road, or birds in a park. Signals were sounds which were
actively listened to such as an announcement over a tannoy
ICAD06 - 106
Proceedings of the 12th International Conference on Auditory Display, London, UK, June 20-23, 2006
system. Soundmark was a derivation of landmark and denoted
a sound unique to the environment such as Edinburgh’s one
o’clock gun. Archetypal represented historical often
“mysterious” sounds such as the creak of ancient wood as it
settled down. Each sound could further be classified as either
centripedal (gathering) or centrifugal (scattering) and the
overall soundscape as either Hi-Fi (High Fidelity) or Lo-Fi
(Low Fidelity). A Hi-Fi soundscape was one where sounds
could be clearly heard against the background, and was usually
accompanied by the ability to hear sounds from a distance.
The Lo-Fi soundscape was normally associated with city life,
where it was difficult to differentiate individual sounds unless
they were amplified.
Having realized that Schaeffer’s [13] sound object
classification only works for “single musical objects” Schafer
developed a greatly reduced system suitable for field notes,
which he enhanced through additional information about the
sound’s setting, estimating distance, intensity, distinctiveness,
ambiance, occurrence, and environmental factors such as
reverb or displacement. His two dimensional notation denotes
attack, body and decay horizontally, and duration, frequency
fluctuations and dynamics vertically. Schafer empirically
generated a catalogue to record information about the
evolution of the soundscape from ‘earwitness accounts’
contained within literature, which was expanded as necessary.
From this he was able to track the gradual change from natural
sounds to those associated with technology, including a
reference to disliking saws by Cicero (c. 70BC). Sonnenschein
adapted this work in order to propose a form of the
classification suitable for the film industry, as there was no
accepted standard [14].
3. SURVEY OF AUDIO PROFESSIONALS’
PRACTICE
In order to survey current practice a twenty-question
questionnaire was e-mailed as an unsolicited word document to
a wide variety of auditory professionals. This was continued
until such time as twenty-five responses had been obtained
from each of the professions in three areas judged as being the
most significant: Acoustics, Computer Science and Design. E-
mail addresses were gleaned from published papers,
membership rolls, newsgroups, and web sites. The response
rate was approximately four percent.
3.1. Participants
Participants were placed into three equal sized groups for
analysis according to their responses about their roles and
responsibilities: Acoustics, Computer Science and Design.
The acoustics group included practitioners in acoustics within
a variety of fields, from building acoustics to psychoacoustics.
The design group included practitioners generally more
concerned with the design and delivery of audio rather than its
measurement. It was found that the designers were rarely
formally trained. The computing practitioners were generally
involved with developing user interfaces incorporating audio,
or writing software to manipulate audio.
Academics formed the largest part of both the Acoustics
and Computing group, while unsurprisingly, the Design group
was essentially comprised of sound designers.
The primary area of work in which respondents were
involved were: Music (13%), Software Development (13%),
Psychoacoustics (12%), HCI (9%), Architectural and Building
Acoustics (7%), Noise and Vibration Acoustics (7%), Theatre
(7%), Games (5%). Other fields represented included Film,
Multimedia, Neuroinformatics, Phonetics, Physics,
Physiology, Technology Development, Television and Radio
in descending order.
Sixty one percent of the participants had been formally
trained, while the remaining 39% attributed their expertise to
industrial experience only. Acousticians had the highest
instance of formal training (76%) with the highest ratio of
PhDs (44%). Designers’ qualifications were predominantly in
music performance or composition.
3.2. Noise
Respondents were asked to provide “definitions of noise and
rank them according to relevance to your [their] field”. A wide
variety of definitions were provided which were subsequently
classified, this provided three clear dimensions which were
shared across all three groups: preference (47%), artefacts
(40%) and spectral (28%). The most common definition was
‘unwanted sound’ (44%), but there was little consensus across
the fields as to a common definition.
3.3. Soundscape
All of the participants understood the concept of the
soundscape, from either the natural or constructed perspective,
but rarely both. One acoustician referenced Schafer [12],
while none made reference to the importance of
psychoacoustics when inhabiting the soundscape. One
acoustician did refer to the importance of the point [of
listening], and range of time. Eighty-eight percent had
encountered the term soundscape with 43% defining it as a
synthesized auditory environment, 33% as the auditory
environment and 21% as the perceived auditory environment,
which is defined in the literature as being the correct definition
[12].
3.4. Description of Aud io
The quantitative and qualitative elements of sound events were
frequently confused, with classifications or descriptions
cropping up in both formal and informal sections of the
questionnaire. Participants’ educational background correlated
positively with the number of quantitative methods used for
measuring sound, as well as the use of formal methods for
classifying sounds. Whilst this threw up a large number of
measurements of sound, which had not been considered in the
research so far, such as ‘coverage’, and ‘clarity’ or
‘intelligibility’, no new forms of classification have been
necessary to date. The most common forms of visualizing
sound were: waveform, spectrogram, time vs. frequency and
musical notation.
The participants employed a greater range of adjectives to
describe sound events than to describe formal measurements,
and these bore a closer relationship to the sounds themselves,
specifically their aesthetics, than to the events that created
them. Each participant was asked to list ten terms, which they
were aware of for describing audio and then rank them
according to their importance within their field. These were
then classified into dimensions and cross-referenced with
frequency and percentage response rate (table 1).
Individual terms were later compared with those of the end
users in order to develop a method of classification that was
meaningful to each group. The three most commonly cited
terms by professionals within dynamics were volume (45%),
ICAD06 - 107
Proceedings of the 12th International Conference on Auditory Display, London, UK, June 20-23, 2006
loudness (24%) and level (12%). In spectral it was pitch
(17%), timbre (15%) and tone (12%). Aesthetics were
brightness (17%), harshness (16%) and warmth (15%). Clarity
was quality (11%), intelligibility (8%) and stereo definition
(7%).
Dimension
Frequency
Response
Dynamics
61
81%
Spectral
44
59%
Aesthetics
38
51%
Clarity
36
48%
Architectural Acoustics
22
29%
Perceptual
20
27%
Type
14
19%
Temporal
13
17%
Reproduction
13
17%
Musical
8
11%
Interacting materials
5
7%
Onomatopoeia
4
5%
Hearing Abilities
2
3%
Table 1: Dimensions which audio professionals were aware of
for describing audio.
3.5. Description of room acoustics
Room acoustics were only fully understood by the
acousticians, and even then, there was a distinct variance.
Non-acousticians often had picked up a few terms, most
commonly ‘reverberation time’ and ‘frequency response’, but
were not familiar with the scales upon which they were
measured using more abstract terms. Sound designers were the
least concerned with the room acoustics, but a couple were
concerned with the reproduction quality of the audio hardware
of end users.
3.6. Summary
Overall there was found to be little overlap of terminology
within the professional fields, except in the most general terms.
There was also little evidence of a desire to notate, classify and
visualize sound events, beyond the standard methods of
waveform and spectrograph. There were specific exceptions
within acoustics, but sound designers and computer scientists
evidenced little need, despite a number of them working on the
auralization of data. However, one of the computing
technologists utilized a very simple, but effective, method of
describing audio: sense of direction; sense of depth; sense of
space; sense of movement; distance to events; broadness;
naturalness; richness; tone colour and emphasis.
Computing participants were comfortable with the term
‘sound event’, whereas sound designers preferred the terms
‘sound’ or ‘audio’, disassociating them from the source. The
overall response to the research varied from not seeing the
relevance, to requesting access to any published results. An
acoustic phonetician suggested that the proposed methods
would prove ideal for use within their field, which they felt
that sound designers and engineers had traditionally ignored.
None of the participants referred to any other researchers
working in this area.
The questionnaire has established the methods and
terminologies audio professionals currently use when notating,
classifying and visualizing sounds. It has confirmed that there
is a wide range of skills and understanding across the fields
closely associated with education, and that many concepts such
as the ‘soundscape’ and ‘noise’ have no standard accepted
definitions, even within the same professional field.
4. HOW END USERS DESCRIBE A SOUNDSCAPE
In order to establish how end users describe a soundscape a
custom eight channel digital audio recording/replay system,
was utilized in order to reproduce the soundfield of the Jack
Kilby Computer Centre (JKCC, main computer lab at Napier
University, 500 seats, 8000 cubic metres) during a typical
afternoon (figure 1). The recording involved eight identical
omni-directional tie-clip microphones, with subsequent
speaker positioning matching the microphones in both floor
position & height. These were positioned into an ellipse at
approximately average ear-height when seated, in order to
emulate the majority of the inhabitants’ positions. Omni-
directional microphones were chosen in order to maximize any
natural reflections as well as to ensure that nothing was “off-
axis” as would be the case with directional microphones.
Figure 1. Picture of the 500 seat computing lab in
which the soundfield was recorded.
The recording was made in a single 30 minute pass onto eight
separate channels, a separate eight channel microphone pre-
amp was used to minimize distortion and ensure consistency in
both dynamics and frequency. Each channel was recorded at
96kHz and 24 bits, which gave us an theoretical dynamic range
of 144 dB and ensured that the full audible range was covered.
Calibration between the physical soundfield and its
subsequent reproduction was achieved utilizing a sound
pressure level (SPL) meter. The meter was set to the C scale
and recorded an average of 48dBC, the A scale would have
rolled off too much bass, whereas the C scale more accurately
represents the acoustic energy present during the recording.
For reproduction eight compact monitors were
supplemented by four sub bass units, whilst bass transmission
can normally be considered omni-directional, the low SPL
levels made accurate positioning of low frequency sounds,
such as people walking on hollow resonant floors, difficult.
The use of four sub bass units resolved this problem, achieving
a more accurate representation, than that normally associated
with a 5.1 or 7.1 system, where sub bass is normally located in
front of the listener. This also compensated for the reduced
frequency transmission range associated with compact
monitors (Figure 2).
ICAD06 - 108
Proceedings of the 12th International Conference on Auditory Display, London, UK, June 20-23, 2006
Figure 2. Representation of the recording/playback
system. 8 microphones were used for recording, 8
speakers and 4 subwoofers were used for playback.
4.1. Experimental protocol
Forty participants were asked to describe the recorded or
natural physical soundfield of the JKCC. Participants’
descriptions were recorded using a standard stereo tie-clip
microphone onto a DAT recorder set to 48kHz 16 bit, this
allowed an accurate stereo image in order to emulate the
participant’s listening experience with reference to their own
voice, as well as a source for later transcription.
The 40 participants were divided into four groups
according to the following conditions:
• Condition 1: while physically present in the JKCC for
15 minutes participants were asked to speak-aloud
what they could hear.
• Condition 2: participants were blindfolded while
physically present in the JKCC for 15 minutes and
were asked to speak-aloud what they could hear.
• Condition 3: participants were exposed to the
recorded soundfield for 15 minutes. They were asked
to speak-aloud what they could hear.
• Condition 4: participants were blindfolded and
exposed to the recorded soundfield for 15 minutes.
They were asked to speak-aloud what they could hear.
The study was conducted over a period of two consecutive
weeks. The participants varied with respect to their age, sex
and background. All participants took part in the study on a
voluntary basis and all were required to have a high command
of spoken English. The use of four different groups reduced
the effect of bias by including subjects who could see sound
sources, as well as those who could only hear them without
any visible clues. The recording allowed half of the groups to
experience an almost identical auditory environment, the
sounds they generated themselves being the only variant. The
physically present groups each experienced a completely
unique environment, which extended the number of auditory
events that could be described by participants.
4.2. Results
Merleau-Ponty’s statement that “it is a matter of describing,
not of explaining or analyzing” accurately represents the
descriptions provided by participants while speaking aloud and
in response to the questionnaire [15].
A number of the participants, who were unaware of where
the recording took place, started by trying to establish what the
space was they were listening to. This initially took the form of
listing the individual sound events and then piecing them
together in order to establish the type of environment. ‘Again
the same sense of people in the distance doing something...
sitting, chatting but all very distance from me say oh... say
fifteen, twenty, thirty feet it does still feel that I'm still in a
large open space but indoors definitely indoors’. This, then
affected their decisions about what sounds they were listening
to.
4.2.1. Themes
A variety of themes arose with the most prevalent being the
source or the “sound of what?” [16], these varied from the
vague ‘somebody’ to the more precise inclusion of gender and
age in ‘young woman’ detailed by only two of the 40
participants. Vocalizations such as ‘speech’, ‘conversation’
and even ‘cough’ formed the largest detailed group, which
corresponds with Cole’s description of children’s preference
for speech over non-speech sounds [17]. Nationality and
accents were identified, together with content, which was
mostly generic ‘saying what’s what’ and ‘asking a question’.
Emotional content was not confined to purely speech, ‘pens
being clicked in frustration’ as well as ‘nervous juggling of
coins in pocket’, four of the participant’s commented on the
poor health of some of the inhabitants of the environment.
When specifying the source of the sound event most
participants were confident of the source even when being
generic. A quarter of the respondents did come across sound
events, which they could not identify, but this represented a
very small amount of the total sound events compared to those,
which they felt they could either estimate or identify.
Comparisons were made, such as the air-conditioning being
‘like the sea coming from behind me’ or ‘a moving airstrip
around me’, but the majority of sources were identified by
single words. Materials where described as being ‘metal’
‘paper’, ‘plastic’ ‘velcro’ or ‘wood’ with the mass described as
either ‘heavy’ or ‘large’ but never light or small.
Actions, which generated the sound source were then
described such as ‘typing’ or the onomatopoeic ‘tapping’.
Individual sound events were generally described only once
until the event varied or a lack of new sources became evident,
at which point the temporal aspect of whether it was ‘constant’
or had just ‘stopped’ were detailed. This varied when applied
to vocalizations, which were mentioned mostly whenever
heard, even from the same source, further reinforcing the
appearance of a predilection for human speech.
Physical properties such as dynamics and spectrum
featured, with the former, despite being mentioned the most,
being mostly confined to ‘loud’ which in turn was translated
into the inferred force of the action such as “hitting the
keyboard hard”. Silence was only mentioned by its absence,
ICAD06 - 109
Proceedings of the 12th International Conference on Auditory Display, London, UK, June 20-23, 2006
which as Cage discovered does not exist outside a vacuum,
even in an anechoic chamber [18]. More interestingly, quiet
sounds were rarely mentioned as being quiet, dynamics were
mostly considered when they became ‘loud’. Spectral aspects
referred mostly to voices with the limitations of ‘deep’ or ‘low’
and the less frequent ‘high’ or ‘higher’.
Clarity was referred to in terms of ‘distinct’ or ‘muffled’
with participants not being able ‘make out’ the speech of the
recordings, which a few found ‘annoying’. Differentiation
between sound sources did occur but more by default rather
than considered identification. Quantities of sound sources
were identified with accuracy between one and four, otherwise
it was a generic ‘few’ or ‘lots’. Only 10% of the participants
referred to whom a sound was directed at, which in all
instances was speech, with a single reference to masking ‘it
drowns out the sound of people talking... well almost...’.
Aesthetics were rarely mentioned those that were, being
mostly negative such as ‘bland’, ‘drone’ and ‘monotonous’,
with spectral aspects referred to as being ‘hard’ or ‘sharp’.
The vast majority of sound source locations were described
in relation to the participant. They were commonly detailed in
terms of left, right front and back with occasional generic
references to distance, ‘I'm starting to recognize the sounds
constantly coming from the top right from my point of view
somebody has just rolled over with their chair along rails in
cluster one...’. A few participants specified height both in the
physical environment and surprisingly on the recording ‘I'm
getting some noise above me to the right...’, which had no
height channel, although this is proposed a future series of
experiments. Individuals were described as ‘walking up and
down steps’ or ‘walking by’, or even moving from ‘left to
right’. Whilst descriptions were always generic they illustrated
an awareness of moving objects rather than a static auditory
environment, ‘there is a bag of crisps flying around... it started
on the front left and then went all the way to the back left...’.
Context was occasionally described in some detail such as
‘I can tell you that someone is pressing the key... and I can
imagine that by the rhythm of their fingers when they press
return or press space’ or as a sequence of events ‘checking of
keys in their pocket in their left pocket... a checking of a
mobile phone... turning it on probably picking up of a bag... of
papers stuffing them in... zipping up the bag’.
The environment itself was described in terms of its size,
‘large’ ‘open plan space’ with two participants guessing the
original location and the others going for either a computer lab
or open plan office. When referring to the physical structure
participants detailed: ‘door’, ‘floor’, ‘grating’, ‘rails’, ‘steps’
with one participant who experienced the unidentified
recording describing a ‘high ceiling’ with ‘plaster walls’.
Echoes were described when establishing the room size with
sound ‘pinging off the pillars’.
Privacy was only considered by a single participant
‘conversation private really...’, whereas pollution, in terms of
distraction and annoyance was more evident ‘it's really quite
annoying actually... I don't particularly like this environment.’
Five of the participants referred to sounds, which they
generated themselves ‘I hear myself talking out loud...’
illustrating how they contribute to their own soundscapes.
Immersion was detailed through comments such as ‘I’m really
beginning to think that I am sitting in the office and not sitting
in a dark room’ and ‘I think if I had eyes I would have turned
around to have a look to see who it was’, which were made by
participants blindfolded listening to the recorded soundfield.
4.2.2. Dimensions
Each term was classified in the same manner as with the audio
professionals, frequency and response rates were then
established allowing a comparison with the results gleaned
from the previous study. Source and actions were clearly the
most common terms utilized when describing the sound events
which the participants heard (table 2). These were both present
in 100% of the responses, with source being mentioned more
often than actions. Sources ranged from a specific reference to
an individual by name through to the more generic ‘bloke’,
which retained gender and quantity, ‘somebody’ was utilized
the most for a single source and ‘people’ for sources which
could not be separated. The use of generic sources such as
‘something’ or ‘keyboard were by far the greatest detailed,
even by the group that could see what the sources were.
Dimension
Frequency
Response
Source
938
100%
Actions
254
100%
Spatial
312
88%
Dynamics
129
80%
Onomatopoeia
117
75%
Temporal
86
73%
Quantity
77
68%
Clarity
43
53%
Comparison
25
48%
Aesthetics
39
45%
Material
36
40%
Spectral
34
35%
Emotions
24
35%
Pollution
17
28%
Architectural Acoustics
10
20%
Table 2: Dimensions which end users utilized when describing
what they could hear.
Spatial dimensions were the third most common, with an
88% response rate, most sound sources being located ‘left’,
‘behind’, ‘right’ and finally ‘front’, with ‘behind’ being almost
twice as common as ‘front’. Spatial aspects were less
important for those who could see and most important for
those who couldn’t, with both blindfolded groups having a
100% response rate compared to 70% and 78% for the sighted
participants.
Surprisingly dynamics such as ‘loud’ and ‘louder’ were
slightly more common than onomatopoeia, here the sighted
groups mentioned it more than the blindfolded, although it was
only referred to a few times by each participant. There were a
wide range of onomatopoeic words, with ‘creaking’ being the
most common, with the blindfolded groups having referred to
it more often than the sighted.
Quantity and temporal dimensions were both very generic,
with ‘lots’ and ‘continuous’ occurring the most. The remaining
dimensions ‘clarity’, ‘material’, ‘spectral’, ‘emotions’,
pollution and ‘arch acoustics’ were rarely referred to when
compared to source and action typically by a factor of
approximately 40:1. However the results do illustrate that
some of the participants were aware of dimensions associated
with musical listening as well as providing an insight into the
terms used. Which in the case of ‘spectral’ were predominantly
‘low’, ‘deep’ or ‘high’. Technical terms such as kHz had no
place in their responses with only a single participant referring
to frequency, and even then only once. Material was not
mentioned at all by the sighted group within the physical
ICAD06 - 110
Proceedings of the 12th International Conference on Auditory Display, London, UK, June 20-23, 2006
environment, and architectural acoustics were only referred to
once by the same group.
Only a single participant, who was listening to the
recording blindfolded, mentioned all of the dimensions with
two more detailing thirteen out of the fourteen. At the other
end of the scale three participants only referred to source and
actions with a third adding spatial references only. Otherwise
participants averaged seven to eight dimensions. Source was
more prevalent for each group over actions.
The results clearly indicate the importance of source and
action when describing sound, but they also show that
participants are aware of other dimensions and have a broad
vocabulary with which to describe them. The use of structured
classification as used when mapping the soundscape should
increase the response beyond the predominant source and
action, with the speak-aloud transcription providing an insight
into how the relevant categories can be expanded.
Clarity, emotions and pollution had not been considered
for mapping, but with a respective response rate of 53%, 35%
and 28%, we are planning to incorporate them into future
experiments. Interestingly, participants only referred to noise
as unidentified sound sources, rather than as an unwanted
sound source or as a sound event without an identifiable pitch.
A trend was noticeable for actions and sounds to be
described indifferently with onomatopoeia. For instance, ‘I
heard a click’ may refer to a click sound or to the clicking
action. This blur may be regarded as a language-dependant
feature which, to the best of the author’s knowledge is quite
sensitive in the English language in comparison to French,
Spanish or Italian, for example.
4.3. Summary
Overall responses varied dramatically in quantity and quality.
The most basic was a series of sound events without sources or
locations, ‘talking… walking… talking… talking… talking…
walking…’. Whereas the other extreme provided rich detailed
information about both the sound sources and their context,
‘Somebody is sitting in front of me and I can hear the typing
quite clearly... he types quite strongly when he used the mouse
I think ... the space on it.’ What is clear though, is that the
information gathered reflects the experience of inhabiting a
soundscape with each individual experiencing as a unique
event, rather than a soundfield, which can only be recorded or
quantified. The act of speaking aloud while inhabiting a
soundscape gives an insight into what Rodaway refers to as
‘the relationship between sense and reality’ [19].
5. DISCUSSION AND CONCLUSION
The two studies presented in this paper involved two clearly
distinct methods, and as such, comparisons between the results
should be carried out with care. Yet, the differences between
the way audio professionals conceptualize, communicate and
represent sounds and the way end users describe it are worthy
of discussion. Lessons can be learnt regarding the following
points: (1) Communication with users about auditory
interfaces. (2) Design of auditory interfaces. (3) Collection of
feedback in user studies.
Given the nature of the speak-aloud study, it was expected
that participants would favor everyday listening when giving
descriptions of their environment. However, what was
unknown was the relative importance of auditory dimensions
based on their descriptions and how this compares to the
auditory dimensions reported by audio professionals (table 3).
Dimension
End User
Response
Audio Pro
Response
Dynamics
80%
81%
Spectral
35%
59%
Aesthetics
45%
51%
Clarity
53%
48%
Architectural Acoustics
20%
29%
Spatial
88%
27%
Temporal
73%
17%
Onomatopoeia
75%
5%
Source
100%
N/A
Actions
100%
N/A
Quantity
68%
N/A
Comparison
48%
N/A
Material
40%
N/A
Emotions
35%
N/A
Pollution
28%
N/A
Perceptual
N/A
27%
Type
N/A
19%
Reproduction
N/A
17%
Musical
N/A
7%
Hearing Abilities
N/A
3%
Table 3: Comparison of dimensions used by end users and
audio professionals.
Overall, if source and actions are discounted, then most
common dimension for both groups was dynamics. The end
users referred to this in terms of high or low whereas the
professionals were more interested in the scale whether it was
volume, loudness or level, when specific terms were used they
mostly related to the upper end of the chosen scale. Spectral
terms were detailed in the same manner, end users again using
high or low, with the professionals concerned with pitch,
timbre, tone and frequency, respectively.
The audio professionals referred to noise as sounds with
particular spectral properties, or related to artifacts, or
unwanted sounds. This contrasts with the way end-users use
the term noise: predominantly to refer to an unidentified
source. Some interesting similarities were also noticed. Both
groups predominantly made aesthetic judgments in negative
terms. This is not surprising for the speak-aloud study, as the
environment participants were asked to describe didn’t offer
much to be aesthetically enthusiastic about. On the other hand,
the fact that professionals submitted a majority of negatively
aesthetic terms suggests that we are more used, or more
effective at experiencing/or at least communicating negative
experiences regarding our auditory environment than positive
ones.
Both groups also described room acoustics in similar
terms, principally referring to reverberation or echo. No
noticeable knowledge gap was noticed in this area. Clarity
judgments were consistently made on a binary scale in both
groups. For example, the professionals used terms such as
rough, smooth, transparent, muffled, dirty, clean. The end
users described events in similar fashion, without any
moderating adverbs. Temporal aspects were referred to
predominately in terms of pace and timing by the
professionals, with the computing specialists being most
concerned. Constancy concerned the end users, whether it
continuous or intermittent.
Both emotions and pollution were referred to by the non –
professional group, but not by the audio pros, this is probably
due to the nature of their work. Practitioners are usually
ICAD06 - 111
Proceedings of the 12th International Conference on Auditory Display, London, UK, June 20-23, 2006
working in acoustic isolation, a sound designer for an interface
does not have to routinely consider the auditory environment
into which their work will be experienced. In contrast end-
users cannot easily isolate themselves to the same degree, even
when using headphones. An understanding of the way in
which end users experience sound pollution and what they
class as pollution would benefit the greatly the design of
auditory interfaces. Often it is only a small element of an
overall design, which commonly can lead to the audio within a
device to be disabled completely.
Emotional responses are a mainstay in music and sound
design for the entertainment industry. But are rarely formally
analyzed, being confined to an individual’s experience.
Emotional content was mentioned by 35% of the end users and
included to varying degrees all of the six basic emotions:
surprise, anger, sadness, disgust, fear and happiness. The
predominant terms related to happiness, followed by fear.
Currently effective auditory design is conducted in
isolation to the end user, and if end user testing is conducted
then trained users are usually preferred in order to
communicate effectively the experience [20]. But when we all
inhabit our soundscapes we don’t think in terms of the
measurable soundfield, we resort to identifying what the
source is and what the action was and where it is coming from.
End users are aware of the dimensions which tax designers,
such as clarity and aesthetics, but they come after the key
dimensions of source and action have been established.
The use of spatial cues is commonly shied away from
during design, due to reproduction problems with accuracy and
the existence of a “sweet spot” [21]. End users described a
more blunt left, right, front, back orientation, which with the
increasing use of HRTF headsets and 5.1 surround sound
systems should encourage auditory professionals to worry less
about the accuracy of reproduction, and experiment more, if
only to increase the apparent clarity due to signal separation
effects.
Among things to investigate further are the differences
between individuals who experienced most of the soundfield as
coming from behind them. Film sound designers have been
aware of this effect for some time and consequently make
sparse use of the rear channels in film soundtracks, referring to
a “sweet spot” after which the surround channels become
“intrusive” [21]. Further study should also be made into the
effect of the participant’s voice when speaking having different
acoustic effects than the recorded soundfield. This can be
rectified by passing the participant’s voice through an
appropriate reverberation unit in order to recreate the effect of
speaking in the environment under study.
The results from this study will be utilized to create a
method of classification for the design and evaluation of
augmented auditory environments, through a technique called
soundscape mapping. The auditory context of use can be
studied by questioning current inhabitants about their
perceived soundscape. This is achieved through classifying
each sound event. The results, after visualization, are then
passed on to the designer for reference purposes. During the
design process the designer will consider what they want the
end-user to experience, and in the process create their own
map using the same classification process. Finally the auditory
elements or interface are studied in situ. or within a simulated
environment, with new maps being created. A comparison of
the subsequent maps will illustrate where the expectations
between designers and end-users match, and what impact the
new elements have on the pre-existing inhabited shared
auditory environment.
6. REFERENCES
[1] B. Delage, "On Sound Design," presented at Hey, Listen!,
Stockholm, Sweden, 1998.
[2] W. Gaver, "What in the word do we hear?" Ecological
Psychology, vol. 5, pp. 1-29, 1993.
[3] C. Macaulay and A. Crerar, "'Observing' the workplace
soundscape: Ethnography and auditory interface design,"
presented at ICAD 98, International Conference on
Auditory Display, University of Glasgow, UK, 1998.
[4] S. A. Brewster Providing a Structured Method for
Integrating Non-Speech Audio into Human-Computer
Interfaces, PhD thesis, Department of Computer Science.
1994, University of York: York.
[5] S. Feld, Sound and Sentiment: Birds, Weeping, Poetics
and Song in Kaluli Expression. 1990, Philadelphia:
University of Pennsylvania Press.
[6] B. Truax, Acoustic Communication. 2nd ed. 2001,
Norwood: Ablex Publishing Corporation.
[7] B. C. J. Moore, An Introduction to the Psychology of
Hearing, 4th ed. ed. London: Academic Press, 1997.
[8] McGregor, I.P., A. Crerar, D. Benyon, and C. Macaulay,
Soundfields and Soundscapes: Reifying Auditory
Communities. ICAD 2002, 2002: p. 290 - 294.
[9] A. Gabrielsson and H. Sjogren, "Perceived sound quality
of sound-reproducing systems," Journal of the Acoustical
Society of America, vol. 65, pp. 1019-1033, 1979.
[10] P. Amphoux, L'identité sonore des villes Europ éennes.
Grenoble/Lausanne: Cresson/IREC, 1997.
[11] B. Hellstrom, "The voice of the place: A case-study of the
soundscape of the city quarter of Klara, Stockholm.," in
Yearbook of Soundscape Studies 'Northern Soundscapes',
vol. 1, R. M. Schafer and H. Jarviluoma, Eds. Tampere:
University of Tampere, Department of Folk Tradition.,
1998, pp. 25-42.
[12] R. M. Schafer, The Tuning of the World. Toronto:
McClelland and Stewart Limited, 1977.
[13] P. Schaeffer, Traité Des Objets Musicaux. Paris: Éditions
du Seuil, 1966.
[14] D. Sonnenschein, Sound Design: The Expressive Power
of Music, Voice and Sound Effects in Cinema: Michael
Wiese Productions, 2001.
[15] M. Merleau-Ponty, Phenomenology of Perception.
London: Routledge, 1962.
[16] C. Metz, "Aural Objects," in Film Sound: Theory and
Practice, E. Weis and J. Belton, Eds. New York:
Columbia University Press, 1985, pp. 154-161.
[17] M. Cole, Cultural Psychology: A Once and Future
Discipline. Cambridge, MA: The Belknap Press of
Harvard University Press, 1996.
[18] J. Cage, Silence: Middletown: Wesleyan University Press,
1961.
[19] P. Rodaway, Sensuous Geographies: Body Sense and
Place. London: Routledge, 1994.
[20] Soderholm, M., Listening Test as a Tool in Sound Quality
Work: Applied to Va cuum Cleaners, The Marcus
Wallenberg Laboratory for Sound and Vibration
Research, Department of Vehicle Engineering. 1998,
Royal Institute of Technology: Stockholm.
[21] W. Murch, "Touch of Silence," in Soundscape: The
School of Sound Lectures 1998-2001, L. Sider, D.
Freeman, and J. Sider, Eds. London: Wallflower Press,
2003, pp. 83-102.
ICAD06 - 112
