Experiment II
Acoustic Gestalt: On the Perceptibility of Melodic Symmetry
The important role of symmetry perception in the visual domain has been well documented in a large number of studies. Less clear, however, is its effect and potential role as a Gestalt grouping principle in the audio domain. We investigated the perceptibility of melodic symmetry using a series of algorithmically generated sonifications. Twenty-eight naïve participants were presented with a series of nine symmetrical sonifications , nine partially symmetrical sonifications (with approximately half of the mirrored elements changed) , and nine asymmetrical sonifications (see graphs below). The participants were asked to identify the sonifications as belonging to one of those three categories. The sonifications utilized Karplus-Strong string synthesis and had a duration between 500 and 8,000 ms. The sonifications were presented three times each in order to check for participants' consistency. Although participants tested far closer to chance level than perfect accuracy, we observed large effect sizes on measures of both accuracy and consistency. We found an effect of the number of tones in a melody on accuracy, with sonifications containing more tones being more difficult to attribute to the correct category. Sonifications with shorter duration and a faster tempo were also found to be more difficult to attribute accurately, indicating some minimum duration of the melody as well as the individual tones constituting the melody. We also found evidence of a significant effect of age on participants’ consistency, with older listeners performing more consistently. The table below shows the results of the experiment. Filters can be used to sort the results. All sonifications used in the experiment can be downloaded here. The results of this study have been published in Musicae Scientiae (2016).

ABOUT US:
Casey Mongoven is a composer specializing in sonification, or the creation of acoustic graphs of mathematical objects. His primary area of research interest is experimental psychology in music and the acoustic domain. More info at caseymongoven.com.

Claus-Christian Carbon is currently head of the Department of General Psychology and Methodology at the Universität Bamberg, in Bamberg, Germany. He is also head of the research group EPӔG (Ergonomics, Psychological Ӕsthetics and Gestaltung). Some of his primary research interests are empirical aesthetics (and design), face recognition/face processing/prosopagnosia (face blindness), optical illusions, and cognitive maps. More info at experimental-psychology.de.

KEY AND INFO ON TABLE:
Accuracy was measured in percent, and given that there are three categories (symmetrical, partially symmetrical, and asymmetrical), if a subject tested at chance level, he or she would score 33.3%. The values reported are the averages of the 81 sonifications presented. Overall accuracy is reported as well as accuracy for each category: AS = accuracy for symmetrical examples; AP = accuracy for partially symmetrical examples; AA = accuracy for asymmetrical examples; A = overall accuracy.

Consistency was measured as 0% for the sonifications presented for which participants chose a different category for each of the three times it was presented, 50% if they chose the same sonification twice, and 100% if they chose the same sonification three times. Thus participants who scored 100% on a single example did not necessarily choose the correct category, but were entirely consistent in their choice for that single example for the three times it was presented. Overall consistency is reported as well as consistency for each category: CS = consistency for symmetrical examples; CP = consistency for partially symmetrical examples; CA = consistency for asymmetrical examples; C = overall consistency.

Musical education (music ed.) was measured on a point scale from 0 to 4. The four questions were: 1) Do you have a degree in music? 2) Do you play an instrument? 3) Do you like to sing? 4) Do you write music? Mathematical education (math ed.) was measured on a point scale from 0 to 2. The two questions were: 1) Do you have a degree in mathematics? 2) Do you often use mathematics in your studies or work? Comments made by the participants and other information can be viewed by rolling over the comment section in the table with the mouse cursor.
Use the filters below to restrict the results. The filter can be used, for example, to limit the duration or note value range for the sonifications presented in the table, given that the sonifications all had a duration between 0.5 and 8.0 seconds, and a note duration between 0.0625 and 1.0 seconds.
Filters
subject no.   from to 1–28
age   from to 8–91
sex  male  female
country of birth  nationality  
Germany
India
Korea
Malaysia
USA
Germany
Malaysia
USA


piece duration   from to s
note duration   from to s
 
Example Sonifications
The sonifications in the experiment were composed algorithmically with pseudo- random integers, such that each participant's set of sonifications was unique. Karplus-Strong synthesis was used used with the harmonic series as a temperament. Participtants were not given visual graphs during actual testing.

Symmetrical Sonification
Partially Symmetrical Sonification
Asymmetrical Sonification
sub.
AS
AP
AA
A
CS
CP
CA
C
sex
age
born in
nat.
music ed.
math ed.
comments
1
63.0%
37.0%
44.4%
48.1%
55.6%
50.0%
66.7%
57.4%
34
Germany
Germany
4
1
lots of this was guess work at best. I thought lower pitched tones gave the most important hints. At times, I was torn between asymmetric and partially symmetric.
2
25.9%
40.7%
25.9%
30.9%
61.1%
66.7%
61.1%
63.0%
44
USA
USA
0
1
Great thanks. Would like the examples in the trial phases to work. Also to have more material to review the differences of each type. Concrete definitions would be best
3
40.7%
33.3%
40.7%
38.3%
66.7%
72.2%
55.6%
64.8%
43
USA
USA
4
0
I have had some ear training and this seemed not to use any of the hearing skills I have practiced. It was frustrating to try to pinpoint where the axis of the split between halves was in the piece. I enjoyed trying to do it, though!
4
66.7%
37.0%
44.4%
49.4%
61.1%
38.9%
44.4%
48.1%
26
USA
USA
3
1
That felt pretty difficult! I don't listen to a lot of microtonal music, so I think a lot of the intervals threw me off. I think some of the easier examples were faster and had dramatic leaps. That way, I was listening more for a rhythmic interplay between bass and treble. I think it's easier for me to listen for symmetric rhythms like that instead of symmetric melodies. If it helps, my primary instrument for years was percussion, so that's the sort of listener that I tend to be.
5
51.9%
33.3%
74.1%
53.1%
50.0%
44.4%
77.8%
57.4%
26
USA
USA
1
1
It would be interesting to look at whether individuals with a background in electronic music perform differently on the test from individuals with a background in classical music, and also how these two groups compare to individuals with no musical background at all.
6
40.7%
22.2%
33.3%
32.1%
61.1%
50.0%
22.2%
44.4%
28
USA
USA
1
0
I was not sure about even one of them. It is hard to have any idea what is going on if you don't know when to expect the middle point, because you don't know what would need to be mirrored, and so your standard for symmetry is essentially hidden from you.
7
55.6%
33.3%
37.0%
42.0%
61.1%
72.2%
38.9%
57.4%
24
USA
USA
0
0
8
88.9%
29.6%
40.7%
53.1%
83.3%
55.6%
55.6%
64.8%
26
India
USA
1
0
9
48.1%
48.1%
33.3%
43.2%
44.4%
50.0%
44.4%
46.3%
22
Korea
USA
2
0
10
44.4%
37.0%
33.3%
38.3%
66.7%
50.0%
50.0%
55.6%
22
USA
USA
1
0
It was great.
11
55.6%
25.9%
44.4%
42.0%
55.6%
50.0%
55.6%
53.7%
18
USA
USA
1
1
Everything kind of sounded the same.
12
33.3%
40.7%
25.9%
33.3%
66.7%
55.6%
50.0%
57.4%
39
USA
USA
4
1
13
48.1%
44.4%
48.1%
46.9%
55.6%
55.6%
72.2%
61.1%
27
USA
USA
1
0
14
63.0%
40.7%
29.6%
44.4%
55.6%
61.1%
61.1%
59.3%
27
USA
USA
3
1
15
48.1%
44.4%
40.7%
44.4%
61.1%
50.0%
55.6%
55.6%
32
Malaysia
Malaysia
1
1
The perception of symmetry in a sequence of notes is arbitrary to the length of notes that we listen to. Therefore, perceiving the center point of symmetry is not a trivial task. Trying to "expect" when the point of reflection is, is somewhat a problem, especially since we humans are used to listening a melodic sequence in the temporal domain, and how one melody relates to the other. Furthermore, our short term memory plays an important role in remembering in a brief time period, for us to perceive what is reflected over the reflection axis. Expecting when the mirroring happens would allow us to better perceive symmetry (perhaps?), but as I understand, that is the whole point of this test.
16
74.1%
33.3%
77.8%
61.7%
66.7%
55.6%
72.2%
64.8%
32
USA
USA
4
1
I found that hearing the symmetry became easier over time. I began picturing the shape of the melodies while they were played, and this seemed to help.
17
70.4%
44.4%
25.9%
46.9%
72.2%
66.7%
44.4%
61.1%
62
USA
USA
2
0
HARD!
18
48.1%
40.7%
37.0%
42.0%
38.9%
44.4%
72.2%
51.9%
64
USA
USA
0
0
The fast sequences were the most difficult for me to determine.
19
44.4%
22.2%
40.7%
35.8%
38.9%
55.6%
61.1%
51.9%
35
USA
USA
1
1
20
25.9%
48.1%
37.0%
37.0%
50.0%
38.9%
44.4%
44.4%
8
USA
USA
1
0
it was fun but hard
21
59.3%
37.0%
37.0%
44.4%
66.7%
66.7%
66.7%
66.7%
86
USA
USA
0
0
Interesting experiment!!! May be somewhat tone-deaf. Had some difficulty differentiatiing between full and partial.
22
48.1%
25.9%
14.8%
29.6%
66.7%
66.7%
44.4%
59.3%
91
USA
USA
0
0
23
44.4%
51.9%
25.9%
40.7%
55.6%
55.6%
61.1%
57.4%
38
USA
USA
4
1
24
48.1%
44.4%
22.2%
38.3%
61.1%
50.0%
44.4%
51.9%
40
Mexico
Mexico
2
0
25
74.1%
29.6%
40.7%
48.1%
61.1%
44.4%
50.0%
51.9%
24
USA
USA
2
0
26
44.4%
48.1%
14.8%
35.8%
50.0%
50.0%
55.6%
51.9%
35
USA
USA
2
1
27
37.0%
40.7%
33.3%
37.0%
50.0%
55.6%
61.1%
55.6%
20
USA
USA
3
0
I often felt like the first two notes in a piece would be a marker for the end of the piece, to check for symmetry. Unfortunately, as the test progressed it became increasingly diffiuclt to rely upon the first two notes.
28
70.4%
40.7%
18.5%
43.2%
55.6%
61.1%
66.7%
61.1%
25
USA
USA
1
0
pressed the wrong answer a couple times other than that it was quite a challene
total
 mean
28
52.2%
37.7%
36.5%
42.2%
58.5%
54.8%
55.6%
56.3%
35.6
p
p
p
p
p
p
p
p
<0.001
0.008
0.250
<0.001
<0.001
0.011
0.024
<0.001