Development of a System to Generate Artiﬁcial Ambiance

based on Entropy

Hidefumi Ohmura

, Takuro Shibayama

2,3

, Masahide Yuasa

, Takayuki Hamano

and Ryu Nakagawa

Department of Information Sciences, Tokyo University of Science, Yamazaki 2641, Noda-shi, Chiba, Japan

Department of Information Systems and Design, Tokyo Denki University, Ishizaka, Hatoyama-cho, Hikigun, Saitama,

Japan

Institute for Music and Acoustics, ZKM, Lorenzstraße 19, D-76135, Karlsruhe, Germany

Department of Applied Computer Sciences, Shonan Institute of Technology, Tsujido Nishikaigan 1–1–25, Fujisawa-shi,

Kanagawa, Japan

Department of Inter media Art, Tokyo University of the Arts, 5000 Omonma, Toride, Ibaraki, Japan

Graduate School of Design and Architecture, Nagoya City University, Chikusa-ku Kitachikusa, 2–1–10, Nagoya-shi, Aichi,

Japan

Keywords:

Ambience, Mood, Emotion, Entropy.

Abstract:

Ambience is an abstract concept of sensory information. If we can control sensory information, then we will

be able to somehow provide ambience as needed. We developed a controllable ambience model using the

entropy of distribution functions based on this hypothesis. We adapted the model to several sensory modes,

such as facial expressions, tones, and unobjective ﬁgures written in HTML and JavaScript available on any

browser. We introduce these systems in this paper.

1 INTRODUCTION

In human communities, there are concepts or words

whose meanings are ambiguous, such as ambience,

emotion, mood, and feeling. But despite their ambi-

guity, we somehow understand them. These words

are used on a daily basis but are deﬁned academically

in some ﬁelds. For example, in psychology, emotion

is deﬁned as rapid mental changes based on obvious

causes. On the other hand, mood is deﬁned as slow

mental changes based on vague causes. Some people

may feel that there are differences between the con-

ventional and academic meanings of words. There-

fore, engineering systems with these academic mean-

ings may not work well for human communities. To

address this problem, we try to deﬁne these concepts

quantitatively in order to maintain their ambiguity.

When the attempt is realized, human-centric applica-

tions will have potential for growth. In this paper, we

focus “ambience” in particular.

We ﬁrst discuss ambience from some perspective.

Based on these perspectives, we deﬁne “ambience”

quantitatively and propose a model for creating artiﬁ-

cial ambiences. Finally, we demonstrate and discuss

applications of the model.

2 DISCUSSION ABOUT

AMBIENCE

2.1 What is Ambience

Ambience is a word with an abstract meaning. Ac-

cording the Oxford Advanced Leaner’s Dictionary,

ambience is “the character and atmosphere of a

place.” Atmosphere is “the feeling or mood that you

have in a particular place or situation; a feeling be-

tween two people or in a group of people.” Feeling is

“something that you feel through the mind or through

the senses,” and mood is “the way a group of people

feel about something; the atmosphere in a place or

among a group of people.” In these sentences, though

we can only perceive that an abstract something con-

trols ambience, we understand that ambience is an in-

terhuman something. Thus, ambience can be consid-

ered as the degree of coincidence of human condi-

tions.

On the other hand, Japanese critic Yamamoto dis-

cussed air (Yamamoto, 1983), which is similar to am-

bience, compared towater, which is similar to a power

of a changing the ambience, during World War II in

Ohmura, H., Shibayama, T., Yuasa, M., Hamano, T. and Nakagawa, R.

Development of a System to Generate Artiﬁcial Ambiance based on Entropy.

DOI: 10.5220/0006730103330338

In Proceedings of the 10th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2018) - Volume 1, pages 333-338

ISBN: 978-989-758-275-2

333

Japan. Members of a community create air. However,

when an authority pours cold water on the air, this

breaks up the air suddenly and creates new air. In his

opinion, we can see that air (ambience) is affected by

the degree of coincidence of members’ something.

2.2 Ambience of Music

As mentioned above, we discussed ambiences as in-

volving two or more people. Here, let us focus ambi-

ence that consists of another medium—music (Lanza,

2004).

In 1934, background music (BGM), referred to as

Muzak, was broadcasted to create a mood or ambi-

ence. BGM is also called elevator music. In the past,

elevators were considered dangerous conveyances

similar to airplanes and roller coasters. Therefore,

BGM was played in elevators to ease passengers’

fears and tension. Music was played not just in el-

evators but also other environments, such as factories,

stores, and so on.

In addition, various musicians created music for

speciﬁc environments. For example, Erik Satie com-

posed “furniture music” in 1920. He created music

that was meant to remain in the background, just like

furniture in rooms. Needless to say, the purpose of

this task was to create and control ambience.

Brian Eno proposed “ambient music” as a musical

genre and composed “Music for Airports” in 1978.

This music melts into the background of airports and

eases the tension felt by passengers before getting on

airplanes.

As shown in the example, there are many chal-

lenges in creating and controlling ambience using mu-

sic. However, there is no well-established methodol-

ogy for creating ambience with music.

2.3 Ambience as an Aggregate of Senses

Ambience depends on various human perceptions.

Ambience is created by both single and multiple

modes of perception. In the previous examples,

modes that perceive the dispositions of members as

well as sounds create ambiences. To give a further

example, the ambience of a painting is created by the

relationships between colors, compositions of objects

in the painting, and/or the artist.

In this study, we deﬁne ambience as an aggregate

of the various aspects of human perception.

2.4 Ambience as Relativity

Ambience is created by the presence or absence of

perception. For example, a situation in which mem-

bers conform to something creates ambience. If a

member opposes this, the situation changes and a new

ambience is created. The reason is because “rela-

tivity” between agreement and disagreement creates

different situations, and members can understand the

difference. Humans are sensitive to change, which

is evident in their physiology. This phenomenon

is observed in electroencephalography (EEG) called

mismatch negativity (MMN) (N¨a¨at¨anen et al., 1978).

EEG is electrical recorded signals in the brain. An

event related potential (ERP) is a particular EEG re-

sponse to a stimulus event. A listener is kept stimu-

lated by sound (e.g., 400 Hz tone for 1 s), and then

abruptly stimulated by another sound (e.g., 600 Hz

tone for 1 s). Comparing the ERPs for the two stim-

uli, we ﬁnd differences at 300 ms after the onset of

the stimulus. This difference, i.e., MMN, shows that

humans continuously search for perceptual deviations

in the environment at both the conscious and uncon-

scious levels.

Although there is relativity of modes, i.e., between

pitches in this example, relativities between modes

create ambience. Let us consider two modes: im-

age and music. Multimedia that consists of a sad im-

age and sad music creates a sad ambience because of

the consistency between modes. On the other hand,

multimedia that consists of a sad image and happy

music creates a curious ambience because of the in-

consistency between modes. Moreover, the relativity

between perceptual modes in music affects ambience

(Ohmura et al., 2013). Complexities of perceptions

in rhythm and melody may be divided into different

musical genres. When both rhythmic and melodic

perceptions are simple, the music sounds like a chil-

dren’s song or nursery rhythm. When rhythmic per-

ception is simple and melodic perception is complex,

the music sounds like jazz. When both rhythmic and

melodic perceptions are complex, the music sounds

like contemporary music. As we have mentioned, we

consider how changes in perceptual mode affect am-

bience.

In this study, we consider that humans perceive

using various senses and pick up ambience from tem-

poral changes and relativities.

3 COMPUTATIONAL AMBIENCE

3.1 Ambience as Deviation

It is said that deviation is one of the important el-

ements of emotion, from theoretical and empirical

points of view. In a musical study, Meyer pointed

out that deviations of expectation arouse emotions

HAMT 2018 - Special Session on Human-centric Applications of Multi-agent Technologies

334

!"#$%&'()*

+&,"-(./01%2&

Figure 1: The optimal complexity model (modiﬁed from

(Berlyne, 1970)).

(Meyer, 1956), based on Dewey’s theory that empha-

sizes the conﬂict or opposition of tendencies (Dewey,

1894). Continuing Meyer’s theory, Narmour de-

ﬁned the relation between expectation and its real-

ization/deviation as the implication-realization theory

(Narmour, 1990). Moreover, Huron proposed an ex-

tended theory known as ITPRA (Huron, 2006).

Although these ﬁndings came from musical study,

when emotional states are regarded as ambiences,

then ambiences depend on deviations.

3.2 Complexity and Ambience

Deviation from an expectation occurs from disappear-

ing or decreasing patterns which are created by hu-

mans. These situations are regarded as increasing

uncertainty and complexity. With regard to com-

plexity, Berlyne proposed the optimized complexity

model (Berlyne, 1970), which shows the relationship

between hedonic emotion and complexity (1). This

relationship is expressed by an inverse U function (x-

axis: complexity, y-axis: hedonic value).

Putting aside the difference between positive and

negative, stimuli that elicit ambiences can be con-

trolled and created by adjusting complexity.

3.3 Entropy and the Information

Theory

An environment where deviations from expectations

occur frequently is regarded as high uncertainty,

and such an environment can be expressed quantita-

tively using information theory (Shannon and Weaver,

1949). Information theory calculates how communi-

cation acts as a function of information.

When an event i occurs, the amount of information

I is deﬁned as

I = −log p

(1)

where p

is the probability of event i. When p

de-

creases, I increases. That is, the amount of informa-

tion increases. When n events occur with the proba-

bilities p

, p

, ..., p

, the expected values are calcu-

lated as

H = −

∑

i=1

log p

(2)

The value H represents the number of times informa-

tion is delivered, i.e., the degree of uncertainty and

complexity. H is the entropy or average amount of

information.

Therefore, by controlling entropy we can create

frequency of deviations, i.e., control artiﬁcial ambi-

ences.

3.4 Meaning and Ambience

In information theory, there are no meanings of

events. For example, giventwo persons and two situa-

tions: in the ﬁrst situation, one person is happy while

the other one is angry; in the second situation, one

person is laughing while the other one is smiling. The

entropies of these two situations have the same val-

ues. That is, information theory dose not distinguish

between “laughing,” “smiling,” and “angry,” and re-

gards them as independent events. In ambience, the

relationships between events are important because

the two examples aboveare not same ambience. In or-

der to distinguish these, we deﬁned the relationships

between events computationally.

Some information on perceptions were deﬁned

from psychological considerations or physiological

features. For example, the relationship of emotion

was proposed as a circumplex model by psycholog-

ical consideration (Russell, 1980). The relationship

of musical notes in music was deﬁned as twelve tones

by physiological features.

When the relationships between some events are

not deﬁned, the designer of the ambience must deﬁne

the relationships between them.

In this study, we develop generative artiﬁcial am-

bience systems based on these considerations.

4 GENERATIVE ARTIFICIAL

AMBIENCE SYSTEMS

4.1 Method for Artiﬁcial Ambience

We proposegenerativeartiﬁcial ambience systems us-

ing one or more perceptual modes. Basically, we de-

ﬁned quantitative relationships and gave each event

Development of a System to Generate Artiﬁcial Ambiance based on Entropy

335

a probability using functions. Functions are deﬁned

based on the normal distribution for these systems. A

number of social and natural phenomena follow nor-

mal distributions; thus, we adopted the normal distri-

bution for the system. We can calculate the probabil-

ities of events using the Gaussian function based on

the normal distribution. The x-axis represents the re-

lationships between events. The y-axis represents the

scale of probabilities. The Gaussian function is repre-

sented as

f(x) =

√

2πσ

exp



−

(x−µ)

2σ



(3)

This equation describes the normal distribution with

average µ and variance σ

. Setting each event to x,

each probability is calculated from µ and σ.

Adjusting the variance σ

of the Gaussian func-

tion, users can control the edge shape of the function.

Upon decreasing the variance σ

, the shape of the

function becomes ﬂat, resembling a uniform distribu-

tion. In this state, the entropy reaches a maximum

that prevents the user from predicting the next event.

Upon increasing the variance σ

, the function peaks.

In this state, the entropy reaches a minimum, allowing

the user to predict the next event easily. Using Gaus-

sian function, we can treat σ

like entropy. We imple-

mented the method using HTML and JavaScript

We describe the details of the systems in the next

section

4.2 Generative Ambience System using

Simple Facial Expressions

We adopted simple facial expressions for the sys-

tem. These facial expressions consist of ﬁgures of

the eyes and mouth. Although the faces are simple,

they can express emotions. The relationship between

emotions was deﬁned as a circumplex model (Rus-

sell, 1980). In this model, emotions were mapped

in two-dimensional space with arousal and valence

as the positive and negative axis, respectively. We

adopted ﬁgures of facial expressions as relationships

in the two-dimensional emotional space in this sys-

tem, as shown in ﬁgure 2.

In this system (ﬁgure 3), when users push the play

button, each face in the 7 × 7 matrix starts to change

depending on its probability at arbitrary time inter-

vals. The y-axis of the distribution function means

circumplex, and both ends can connect. Arbitrary

time intervals are set with Interval slider.

http://sites.google.com/site/hidefumiohmura/home/

program/icaart2018

!"##$!

%&''&() *"+,)

-./'$!

%+00#)

-'&12"+!

3+0"2".4!5.#+0"2".4)

Figure 2: Relationships between simple ﬁgures of facial ex-

pressions.

Figure 3: Generative ambience system using simple facial

expressions.

4.3 Generative Ambience System using

Sounds

We adopted sine waves as tones to the system. We

adopted the physical ratio of frequencyas the relation-

ship. The relationships of double (2:1) or half (1:2)

in frequency are called octaves, the relationships of

3:2 or 3:4 are called the perfect ﬁfth, and the relation-

ships of 4:3 or 2:3 are called the perfect fourth. Using

these relationships and ignoring octaves, we can get

twelve notes as a circumplex model. The relationship

is called the circle of ﬁfths or fourths. Clockwise ro-

tation means 3/2 × n, anti-clockwise rotation means

2/3 × n, where n is an arbitrary frequency. In this

system, the range of frequency is in four octaves.

When a user pushes the play button (see ﬁgure 4),

the system produces sound that depends on the model

at arbitrary time intervals. Arbitrary time intervals are

set with Interval slider. Note that the values are set

with Value slider.

HAMT 2018 - Special Session on Human-centric Applications of Multi-agent Technologies

336

Figure 4: Generative ambience system using sounds.

!"!

#!"!

$%&"!

'&"!

Figure 5: Relationships between abstract ﬁgures.

4.4 Generative Ambience System using

Abstract Figures

In order to get rid of meaning, we adopted abstract

ﬁgures to the system (ﬁgure 6). Basically, this system

is similar to the system using simple facial expres-

sions. However, the ﬁgures are lines in circles that do

not express meaning. The relationships between these

ﬁgures are angles (see ﬁgure 5).

Figure 6: Generative ambience system using abstract ﬁg-

ures.

Figure 7: Generative ambience system using sounds and

abstract ﬁgures.

4.5 Generative Ambience System using

Sounds and Abstract Figures

In order to express multiple modes, we adopted

both sounds and abstract ﬁgures to the system (ﬁg-

ure 7). This system creates ambience based on

the comparison between entropies of sounds and ab-

stract ﬁgures, and generates them synchronously or

asynchronously. Abstract ﬁgures: lines are con-

trolled by three modes—angle, x-position, and y-

position. Sounds: pitches are controlled by a percep-

tual mode—pitches.

4.6 Discussion

Using each system, we ﬁnd that ambiences change

depending on the entropies. The system using sim-

ple facial expressions generates concrete meanings.

However, when the entropy is high, we cannot un-

derstand the concrete meanings. We really consider

that phenomenon as ambience. The other systems do

not have concrete meanings. Therefore, in order to

understand ambience, we need to compare other out-

puts. It is not clear how concrete/abstract meanings

affect ambience. It will be interesting to determine

what kinds of ambience other distribution functions

create. We plan to conduct experiments on these in

the future.

The elements of these systems, such as sounds

and ﬁgures, are simple. When we adopt complex ele-

ments, such as agents to the system, we may develop

a more interesting system. We will try this in future

work for a multi-agent system.

Development of a System to Generate Artiﬁcial Ambiance based on Entropy

337

5 CONCLUSION

In this study, we deﬁned ambience as an aggregate

of various senses of human perception. We devel-

oped generative ambience systems using these rela-

tionships and a normal distribution based on informa-

tion entropy. This system created ambience with mul-

timodal perception, such as sound, faces, and ﬁgures.

We plan to conduct experiments for the various ambi-

ences in the future.

ACKNOWLEDGEMENTS

This work was supported by JSPS KAKENHI Grant

Numbers JP17K12808, JP16H01744, JP15H03175.

REFERENCES

Berlyne, D. E. (1970). Novelty, complexity, and hedonic

value. Perception and Psychophysics, 8(5):279–286.

Dewey, J. (1894). The theory of emotion: I: Emotional atti-

tude. Psychological Review, 1(6):553–569.

Huron, D. (2006). Sweet anticipation: Music and the psy-

chology of expectation. MIT Press, Cambridge.

Lanza, J. (2004). Elevator Music: A Surreal History of

Muzak, Easy-listening, and Other Moodsong. St. Mar-

tin’s Press, New York.

Meyer, L. B. (1956). Emotion and meaning in music. Uni-

versity of Chicago Press, Chicago.

N¨a¨at¨anen, R., Gaillard, A. W. K., and M¨antysalo, S. (1978).

Early selective-attention effect on evoked potential

reinterpreted. Psychophysiology, 30(5):436–450.

Narmour, E. (1990). Emotion and meaning in music. Uni-

versity of Chicago Press, Chicago.

Ohmura, H., Shibayama, T., Shibuya, T., Takahashi, T.,

Okanoya, K., and Furukawa, K. (2013). Modeling

of melodic rhythm based on entropy toward creating

expectation and emotion. Proceeding of Sound and

Music Computing 2013 (SMC), pages 67–73.

Russell, J. A.(1980). A circumplex model of affect. Journal

of Personality and Social Psychology, 39:1161–1178.

Shannon, C. E. and Weaver, W. (1949). The Mathemati-

cal Theory of Communication. University of Illinois

Press, Illinois.

Yamamoto, S. (1983). Kuki no Kenkyu (Study of Air).

Bungeishuju, Tokyo. (Japanease).

HAMT 2018 - Special Session on Human-centric Applications of Multi-agent Technologies

338