Вы находитесь на странице: 1из 11

Running

head: EMOTIONAL CONTEXT OF MUSIC COMPOSED BY HUMANS VS


COMPUTERS










Emotional context of music produced by humans


versus computer-generated music.
J.A.H.M. van Abeelen, D.A.E. Bonn, R.J.L. van Hamond, Y.M.F.M. Nijs, C. Roex
Tilburg University Human Aspects of Information Technology

EMOTIONAL CONTEXT OF MUSIC COMPOSED BY HUMANS VS COMPUTERS

ABSTRACT
Computer-composed music is becoming a major key instrument to measure the overall
capabilities of software and is potentially hard to distinguish with human composed music.
ANTON 2.0 is such a computer system that is able to generate music using mathematical
formulas, which take melody, harmony and rhythm into account. The focus of this research
lies in the emotional content of computer-composed music.
To find answers, participants in our experiment had to listen to two music samples and
answer questions in an online survey. These questions measured the levels of valence, arousal
and emotion people experienced after listening to a music sample. Other questions wanted to
pinpoint if people are able to determine if the second music sample was human composed or
computer generated. The participants were divided in four conditions. Two conditions
concerned computer-composed music and two human-composed music.
A significant difference was found between the computer-composed and humancomposed conditions. The computer-composed condition scored higher on emotion.
Approximately 70% of the people also answered correct whether they were listening to
human-composed or computer-composed music. The main question if computer-composed
music would assign some or even the same emotions as similar human-composed music was
confirmed. Conflicting differences were found, but still present and therefore this question
was confirmatory.
The main question if computer-composed music would assign some or even the same
emotions as similar human-composed music was confirmed. Noticeable, contradictions were
found: Computer-composed was rated higher on valence and arousal where human-composed
was rated higher at emotion. Conceivably, the origin of this contradiction can be derived from
the fact that the joviality of the music fragments can differ.

INTRODUCTION

and are partly through this real compared

Could computer-composed music assign

to human-composed music, suggested

some or even the same emotions as similar

professional musicians (Boenn, Brain, De

music composed by humans? This thought

Vos & Ffitch, 2011).

is retrieved in ANTON 2.0, a computer

According to Vstfjll et al. (2008)

program that can generate music with

music has some influence in how the

melody,

by

listener felt. Computers are able to

music

generate music from scratch (Wallis,

fragments by ANTON sound quite realistic

Ingalls, Campana & Goodman, 2011)

harmony

mathematical

and

formulas.

rhythm
The

EMOTIONAL CONTEXT OF MUSIC COMPOSED BY HUMANS VS COMPUTERS

based on algorithms. People feel emotions

In this research, conducted by five students

while they are listening to this kind of

of Tilburg University, computer-composed

music, the study found. There is also a

and

correlation between arousal (the degree of

examined to determine whether people rate

activation) and the computer-generated

this kind of music different. The purpose

music. This evidence is in contradiction to

was twofold. Firstly, whether music by

the notion that computers are not able to

computers might result in emotions. And

raise emotions (Sufeng, 2011).

secondly, whether this music may lead to

human-composed

music

were

However, the question is whether

the same emotions as human-composed

people give emotions the same rating as

music causes. The participants were

music generated by humans. Wallis et al.

divided in four conditions. Two conditions

(2011) did not compare their findings to

concerned computer-composed music and

human music, while this is important for

two

further use. For instance, only when people

conditions

give

and

music in which was told that a human

computer-music the same rating, it is

made the music (False belief Human;

possible to determine which computer

FbH), computer-composed wherein the

systems are already able to create real

computer made the music (True belief

humanlike music. This could be useful in a

Computer; TbC), human-composed music

variety of uses. For example, in a game the

in which was told the human was the

computer could generate and play different

componist (True belief Human; TbH) and

songs depending on the level the player is

finally human-composed music whereby

in. The sound would be adjustable to the

the computer made the music (False belief

behaviour of the player in the game too.

Computer; FbC).

samples

of

human-music

human-composed
were

music.

The

computer-composed

Supper (2001) suggested that algorithmic


composition might lead to numerous
musical

compositions.

Therefore

the

possibilities of computer-composed music


seem

limitless.

algorithmic
implemented,

However,

composition
it

is

before

could

necessary

be
that

computer-composed music may result in


the same emotions as human music.
Otherwise this kind of use is superfluous.

HYPOTHESIS
In this study, we will examine whether
people will attribute/assign emotions to
computer-generated music. The hypotheses
are formulated thus:
H1: Will people find it easier to assign
emotions to a song when they
believe a human produced it?
H2: Will people evaluate music as better

EMOTIONAL CONTEXT OF MUSIC COMPOSED BY HUMANS VS COMPUTERS


when

they

believe

humans

to gain the participants thoughts of the

produced it, even though it was

different

actually computer-composed?

questionnaire, there were questions about

H3: Are people able to pinpoint whether it


is

composition

was

music

pieces.

In

the

the evaluation of the music (I enjoyed

human-

listening to this piece of music), which

composed or computer-composed?

had to be answered on a seven-point Likert


scale with seven categories, from strongly

METHOD

disagree, up to strongly agree. The

Participants

participants were also asked if they

The participants were relatives, friends and

believed

acquaintances of students from Tilburg

There was also a question where the

University. There was no focus on gender

participants had to indicate whether they

difference or age difference in this study,

believed

so these factors were not relevant in the

composed

selection of participants. The questionnaire

question concerned a different sample than

was posted online. In total there were 91

the first questions.

the

music-evoked

the

sample

or

was

emotions.

computer-

human-composed.

This

participants, 25 in the FbH condition, 22 in

The computer-composed music was

the TbC condition, 20 in the TbH condition

created by an automated system, named

and 24 in the FbC condition. The average

ANTON 2.0 (Boenn, Brain, De Vos &

age of the participants in the experiment

Ffitch 2011). The most natural sounding

was 25.8 years (SD = 11.4). 57 participants

pieces created by ANTON 2.0 were chosen

were male (avg. age = 24.4 years, SD

for the experiment.

=10.4) and 34 participants were female

The

(avg. age = 28.0 years, SD = 12.6).

human-composed

music were classical pieces chosen for the


experiment were chosen based on their

Materials

similarity to the pieces created by ANTON

Various materials were used to measure

2.0,

the evaluation of computer-composed and

participant would recognize the piece. The

human-composed music. The stimulus

pieces were .wav format samples, taken

material consisted of YouTube videos that

from the website of an Italian label:

played a .wav sample of either a piece of

OnClassical

computer-composed music or a piece of

(http://www.onclassical.com/).

human-composed music.
An online questionnaire was used

and

their

unpopularity-

so

no

All four conditions contained the


exact same questions. The introductions

EMOTIONAL CONTEXT OF MUSIC COMPOSED BY HUMANS VS COMPUTERS

varied: Two conditions were introduced

to indicate whether they believed the piece

with the sentence You are going to listen

was

to a piece of music that was composed by a

composed and to what extent they were

human composer.

certain of their answer.

conditions

were

The other two

introduced

with

human-composed

or

computer-

the

sentence You are going to listen to a piece

Design

of

The experiment consisted of the following

computer-composed

implies

that

in

two

music.

This

conditions,

the

participants were deceived, since the


samples were the opposite.

four conditions:
1. Expose

computer-composed

music, say that it was created by a


human

Procedure

composer

(False

belief

Human; FbH)

Prior to the evaluation of the music, the


participants

to

were

asked

some

2. Expose

to

computer-composed

basic

music, say that it was created by a

questions about their age and gender.

computer (True belief Computer;

Subsequently, the participants were told to

TbC)

listen to a piece of music, which differed

3. Expose to music created by a

per condition: The computer-composed

human composer, say that it was

music fragment or the human composition.

created by a human composer

They were allowed to listen to the

(True belief Human; TbH)

fragment multiple times. Subsequently, the

4. Expose to music created by a

participants were asked to evaluate the

human composer, say that it was

sample and answer questions about the

created by a computer (False belief

emotions, valence and arousal. Finally, the

Computer; FbC)

participants were asked to listen to another

Table 1 provides a clear overview of the

fragment that was either human-composed

conditions.

or computer-composed. They were invited

EMOTIONAL CONTEXT OF MUSIC COMPOSED BY HUMANS VS COMPUTERS


Table 1

Overview of the four conditions in the experiment. Informed composer is what

the participants were told. Actual composer is the music they were actually exposed to
Informed composer
Computer

Human

Computer

True belief computer (TbC)

False belief human (FbH)

Human

False belief computer (FbC)

True belief human (TbH)

Actual composer

The independent variables in this study

composed music and human-composed

were the music (computer-composed or

music

human-composed) and the information

or
Another

not.

confound

was

the

given to the participants before the

possibility that a participant already knew

experiment, whether they were told a

a classical piece, but that they were told

human composer made the music or that it

that it was computer-composed. If this was

was computer-composed (true/false belief

the case, the participants knowledge of

human/computer).

classical music influenced the outcome of

The dependent variables in this

their experiment. However, this was

study were the evaluations of the samples,

unlikely to happen, because of the short

the ease with which the participants

interval of the music sample and different

indicated they could pick an emotion that

(lower) quality of the original music piece.

the song made them feel, and if they could


pinpoint whether a sample was computer-

RESULTS

composed

This research has been conducted to find

or

human-composed.

Since the last research question


concerned

the

ability

of

people

to

potential

differences

emotions

in

between

human

computer-composed

recognize whether a piece is human-

human-composed

composed or computer-composed, this

questionnaire that was used consisted out

could be called the control group, because

of

the participants were asked to tell what

composed music and human-composed

kind of music it is and indicate how certain

music. These two conditions were split

they were. This shows whether they are

into two sub-conditions, in which the

able to distinguish between computer-

participants were either told that the music

two

main

music.

and

conditions:

The
computer-

EMOTIONAL CONTEXT OF MUSIC COMPOSED BY HUMANS VS COMPUTERS


was

computer-composed

or

human-

composed. This made a total of four

between the scores on the question about


emotional charge (t(88)=0.15, p=.88).

conditions. The results will be discussed

Table 3 shows the score on

for each hypothesis.

perceived

emotion

for

the

human-

H1: Will people find it easier to assign

composed and the computer-composed

emotions to a song when they believe a

condition, regardless of what sub-condition

human produced it?

the participants were in. An independent t-

We asked the participant to what

test proves that there is a significant

extent he or she had experienced emotions

difference between the actual computer-

after listening to the audio file. An

composed and actual human-composed

independent t-test was used to compare the

condition (t(88)=5.98, p<.0001), effect-

human belief conditions (TbH, FbH) with

consistency = 27.9%.

the computer belief conditions (TbC, FbC).


There

was

no

significant

difference

Table 3 Score on emotion for the human-composed conditions and computer-composed


conditions (minimum score 1 and maximum score 7; standard deviation between brackets)
Computer-composed (TbC, FbH)

Human-composed (TbH, FbC)

2.72 (1.34)

4.52 (1.52)

Score on perceived emotion

Furthermore,

there

were

questions

composed and the computer-composed

concerning valence and arousal (Oliveira

conditions,

& Cardoso, 2008). The valence & arousal

condition the participants were in. As for

subscale consisted of 2 items ( = .82). For

table 3, there was a significant difference

valence & arousal, there was no significant

between the human-composed conditions

difference between the human belief

and the computer-composed conditions

conditions (TbH, FbH) and the computer

(t(88)=3.07, p<.005). Effect consistency =

belief conditions (TbC, FbC) (t(88)=1.16,

8.7%. The computer-composed conditions

p=.25).

scored higher than the human-composed


Table 4 shows the scores on

valence

&

arousal

for

the

human-

condition.

regardless

of

what

sub-

EMOTIONAL CONTEXT OF MUSIC COMPOSED BY HUMANS VS COMPUTERS

Table 4 Score on valence & arousal for the human-composed conditions and computercomposed conditions (minimum score 1 and maximum score 7; standard deviation between
brackets)
Computer-composed (TbC, FbH)

Human-composed (TbH, FbC)

3.21 (1.51)

2.34 (1.11)

Score on valence & arousal

H2:

Will people evaluate music as

There was no significant difference on

better when they believe humans produced

liking between the true belief conditions

it, even though it was actually computer-

(t(40)=0.12, p=.91).

composed?
An

independent

t-test

was

H3:

Are people able to pinpoint whether

performed on the scores on liking between

it is composition was human-composed or

the computer-composed conditions (FbH,

computer-composed?

TbC). There was no significant difference

The participants were asked to tell

between the scores on liking within the

who or what created the music they were

computer-composed

conditions

exposed to. 72.5% of the participants knew

(t(44)=1.26, p=.22). Another independent

that the music in the first condition was

t-test was performed on the scores on

human-composed.

liking

human-composed

participants knew that the music in the

conditions (FbC, TbH). There was no

second condition was computer-composed.

difference in liking between the human-

In total, participants were more often

composed conditions (t(42)=0.48, p=.64).

correct than incorrect when answering the

between

An

the

independent

t-test

69.6%

of

the

was

question who or what created the music

performed to find out if there was a

((1)=23.42, p<.0001). Table 8 shows the

difference between scores on liking for the

amount of participants that were correct

true belief conditions (TbC and TbH).

and incorrect when answering the question.

Table 8 Confusion matrix of all correct and incorrect guesses on the composer question.
(Scores are the amount of participants. The expected amount is between brackets; N=87)
Guessed
Computer

Human

Actual

Computer

32 (20.7)

12 (23.3)

Composed

Human

9 (20.3)

34 (22.7)

EMOTIONAL CONTEXT OF MUSIC COMPOSED BY HUMANS VS COMPUTERS

CONCLUSION AND DISCUSSION

Furthermore, this finding is additional to

This

potential

the study of Wallis et al. (2011), which

differences of emotional load in computer-

found that people feel emotions while they

composed and human-composed music.

are listening to music, while Wallis et al.

An online survey was used to collect data

did not investigate possible differences

for this purpose. Participants heard various

between the composers.

article

is

about

the

kinds of music depending on the condition

Although we did not find a

they were in. There were four conditions,

difference in this study between the human

from which two concerned computer-

belief and computer belief conditions on

composed

music

human-

emotion, there is one on emotions in the

composed

music:

computer-composed

computer-composed and human-composed

and

two

music in which was told that a human

conditions.

Therefore

made the music (False belief Human;

suggests that there is a discrepancy

FbH), computer-composed wherein the

between

computer made the music (True belief

questionnaire received different questions

Computer; TbC), human-composed music

for valence and arousal and a separate

in which was told the human was the

question for perceived emotion. The

composer (True belief Human; TbH) and

human-composed

finally human-composed music whereby

higher as regarding perceived emotion,

the computer made the music (False belief

while the computer-composed fragment

Computer; FbC).

was rated higher when looking at valence

the

music

the

evidence

fragments.

fragment

was

The

rated

According to the first hypothesis

and arousal. This is conflicting, because

(H1: Will people find it easier to assign

valence and arousal should be proportional

emotions to a song when they believe a

to

human produced it?), people find it easier

consists of valence and arousal. This could

to assign emotions to a song when they

be explained by observing the music

believe a human produced it. In reviewing

fragments, because in this study, no pre-

the results, no data was found on the

research has been conducted on the

association between emotions and belief.

fragments themselves. Participants did not

The results of this investigation show that

rate the fragments, whereby it is unclear

people do not mind what type of composer

what their opinion is regarding various

the music created in their process of

composers, although music fragments

assigning

could have some important differences

emotions

to

the

music.

perceived

emotion

since

emotion

EMOTIONAL CONTEXT OF MUSIC COMPOSED BY HUMANS VS COMPUTERS

10

(Oliveira & Cardoso, 2008). Therefore it

computers cannot make music as good as

could be useful for future research to add

humans, it might be possible that they rate

pre-research to make the study more

computer-composed

reliable.

Furthermore, a within-subject design could

music

lower.

Secondly, according to the second

be used in future research to determine

hypothesis (H2: Will people evaluate

whether people adjust their liking on the

music as better when they believe humans

music composer. For instance, participants

produced it, even though it was actually

could be asked how they would rate a

computer-composed?),

will

music fragment when first is told a human

evaluate music as better when they believe

composed it and when is told it was

humans produced it, even though it was

composed by a computer and vice versa.

actually computer-composed. However, no

Finally, an effect was found according to

difference was found when comparing the

the last hypothesis (H3: Are people able to

false belief human (FbH) and true belief

pinpoint whether it is composition was

computer (TbC), the false belief Computer

human-composed

(FbC) and true belief computer (TbH) and

composed?), which stated that people

the TbC and TbH conditions. While it was

could pinpoint whether a composition is

expected

computer-

human-composed or computer-composed.

composed music lower than human-

Around seventy per cent of our participants

composed music, the results suggest that

could identify the composer. This means it

people do not mind the type of music

might be clear to most people who the

composer when they are evaluating the

music composer was. It might be possible

music. This effect was possibly not found

that people knew that they were fooled in

because

about

the earlier questions for the first two

However,

hypotheses and that they anticipated on

participants adjusted their opinion on the

this. For future research it is recommended

maker of the music and therefore they

to ask participants afterwards whether they

could anticipate on the fact that the music

thought they were fooled.

that

people

people

people

computer-composed

think

rate

worse

music.

or

computer-

was made by a computer. Probably, the

Summarizing, computer-composed

participants thought computer-composed

music evokes some or even the same

music was better than expected. In future

emotions as similar music composed by

research, participants could be asked to

humans, this study shows. Significant

give an opinion about what they thought of

differences were found concerning the

the music they heard. If people think that

assignment

of

emotions

to

music.

EMOTIONAL CONTEXT OF MUSIC COMPOSED BY HUMANS VS COMPUTERS

11

Interestingly, the results of valence and

by desired valence and arousal.

arousal, which are belonging to emotion,

Paper presented at the Proceedings

were different than the results of the

of 8th international sound and

perceived emotion.

music

computing

Padova, Italy
BIBLIOGRAPHY
Boenn, G., Brain, M., De Vos, M., &
Ffitch, J. (2011). Automatic music
composition using answer set
programming.

Theory

and

Practice of Logic Programming,


11, 397-427
Juslin, P., Liljestro m, S., Vstfjll, D.,
Barradas, G. & Silva, A. (2008).
An Experience Sampling Study of
Emotional Reactions to Music:
Listener, Music, and Situation.
Emotion, 8(5), 668683
Oliveira, A., & Cardoso, A. (2008, July).
Modeling affective content of
music: a knowledge base approach.
In Sound and Music Computing
Conference.
Sufeng, Y. (2011). Computer-Assisted
Language

Learning

in

higher

education. IT in Medicine and


Education, 1, 137-141
Supper, M. (2001). A Few Remarks on
Algorithmic

Composition.

Computer Music Journal, 25:1, 4853


Wallis, I., Ingalls, T., Campana, E., &
Goodman, J. (2011). A rule-based
generative music system controlled

conference,

Вам также может понравиться