Melodic Segmentation Across Cultural Traditions M ARCELO E. R - - PowerPoint PPT Presentation

▶
melodic segmentation across cultural traditions
SMART_READER_LITE
LIVE PREVIEW

Melodic Segmentation Across Cultural Traditions M ARCELO E. R - - PowerPoint PPT Presentation

May 18, 2023 581 likes •966 views

Melodic Segmentation Across Cultural Traditions M ARCELO E. R ODRGUEZ -L PEZ & A NJA V OLK Department of Information and Computing Sciences Utrecht University, Interaction Technology Group

slide-1
SLIDE 1

Melodic Segmentation Across Cultural Traditions

MARCELO E. RODRÍGUEZ-LÓPEZ & ANJA VOLK

Department of Information and Computing Sciences Utrecht University, Interaction Technology Group ▼✳❊✳❘♦❞r✐❣✉❡③▲♦♣❡③❅✉✉✳♥❧ June 13, 2014

slide-2
SLIDE 2

2/38

What is Segmentation

slide-3
SLIDE 3

3/38

What is Segmentation

Segmenting = Chunking = Grouping psychology: process of grouping individual units of information into larger units. computing: process of dividing data into smaller meaningful units.

slide-4
SLIDE 4

4/38

What is Segmentation

Segmenting = Chunking = Grouping Psychology: process of grouping individual units of information into larger units. computing: process of dividing data into smaller meaningful units.

slide-5
SLIDE 5

5/38

What is Segmentation

Segmenting = Chunking = Grouping Psychology: process of grouping individual units of information into larger units. Computing: process of dividing data into smaller meaningful units.

slide-6
SLIDE 6

6/38

Music Segmentation: DM Task Definition

In Digital Musicology (DM), segmentation is the task

  • f dividing a musical piece/melody/section into

‘smaller’ structural units.

slide-7
SLIDE 7

7/38

Music Segmentation: Scope

◮ Type of Music and Segment Granularity

  • input: monophonic music (most generally melodies)
  • encoding: symbolic
  • music-theoretic segment parallel: phrases
slide-8
SLIDE 8

8/38

Melody Segmentation: an Example

Aihu renmin zidibing

e

H

q q q q q q q q q q q q

Q Q Q Q EQ E Q

Q q q q.

2

&# 4

##

slide-9
SLIDE 9

9/38

Melody Segmentation: Task

◮ Task

  • identify segment boundary locations
  • pair boundaries (begin, end)
  • label boundaries
slide-10
SLIDE 10

10/38

Melody Segmentation: Task

e

H

q q q q q q q q q q q q

Q Q Q Q EQ E Q

Q q q q.

2

&# 4

##

phrase boundary

slide-11
SLIDE 11

11/38

Models of Melody Segmentation

  • development: +30 years
  • # of models: ≈27
  • comparative studies: 4, 3-6 models evaluated
  • most successful: Gestalt-based
slide-12
SLIDE 12

12/38

Gestalt Models: Basic Overview

  • attempt to quantify (visual) Gestalt principles
  • use system of preference rules
  • Gestalt proximity is modeled as ‘discontinuity’

detection (breaks in the melodic flow)

slide-13
SLIDE 13

13/38

Local Discontinuities Detection: an Example

slide-14
SLIDE 14

14/38

Local Discontinuities Detection: an Example

H

q q q q q q q q q q q q e

Q Q Q Q EQ E Q

Q q q q.

slide-15
SLIDE 15

15/38

Local Discontinuities Detection: an Example

e

H

q q q q q q q q q q q q

Q Q Q Q EQ E Q

Q q q q.

slide-16
SLIDE 16

16/38

Local Discontinuities Detection: Input

e

H

q q q q q q q q q q q q

Q Q Q Q EQ E Q

Q q q q.

e1

. . .

ei

. . .

eN

24 48 72 84 96 120 2322 69 69 68 66 66 69 64

. . . . . . . . . . . .

pitch:

  • nset:
slide-17
SLIDE 17

17/38

Local Discontinuities Detection: Input

e

H

q q q q q q q q q q q q

Q Q Q Q EQ E Q

Q q q q.

.5 .5 .5 .25 .25 .5 .5 1 .5 .5 .5 .5 2 .75 .25 .5 .5 .5 1 .5 .5 .5 .5 .5 .5 0 1 2 0 3 5 0 5 1 2 3 2 5 2 2 3 3 5 2 3 7 1 2 3 5

p-inv : ioi :

slide-18
SLIDE 18

18/38

Local Discontinuity Detection: Profiles

e

H

q q q q q q q q q q q q

Q Q Q Q EQ E Q

Q q q q.

p-inv : ioi :

slide-19
SLIDE 19

19/38

Local Discontinuity Detection: Profiles

e

H

q q q q q q q q q q q q

Q Q Q Q EQ E Q

Q q q q.

p-inv : ioi :

slide-20
SLIDE 20

20/38

Local Discontinuity Detection: Output

e

H

q q q q q q q q q q q q

Q Q Q Q EQ E Q

Q q q q.

0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0

Output :

slide-21
SLIDE 21

21/38

Gestalt Models: Assumptions

  • discontinuity as a distance in some perceptual space
  • discontinuity can be treated as a local phenomenon
  • discontinuity is universal/idiom-independent
slide-22
SLIDE 22

22/38

Statistical Study: Research Questions

  • relevance of discontinuity as a boundary cue?
  • relevance of discontinuity parametric representation?
  • appropriateness of locality for modeling?
  • appropriateness of universality assumption?
slide-23
SLIDE 23

23/38

Statistical Study: Data

ESSEN FOLK SONG COLLECTION (EFSC)

  • 20,000 annotated vocal folk melodies
  • Representative subsets of German and Chinese songs

EXPERIMENTAL SUBSETS

  • ≈ 5k German songs, ≈ 30k phrases, 1 . . . 29 notes
  • ≈ 2k Chinese songs, ≈ 12k phrases, 1 . . . 169 notes

FILTERED

  • ≈ 1.2k German songs, ≈ 5.5k phr-pairs, 13 . . . 21 notes
  • ≈ 1.4k Chinese songs, ≈ 4.6k phr-pairs, 2 . . . 28 notes
slide-24
SLIDE 24

24/38

Statistical Study: Processing

e

2

&# H

4 q q q q q q q q

q q q q

Q Q Q Q EQ E Q

Q q q q.

##

ph a ph b c(ph a) j(ph ab) c(ph b) p-inv : Nioi :

.35 .5 .15 2 5 2

slide-25
SLIDE 25

25/38

Results 1:

pitch duration Germany China d

  • 32.41 %

43.73 %

  • d

49.57 % 72.56 % d

  • 04.51 %

22.53 %

  • d

24.13 % 57.70 % nd

  • 31.49 %

28.53 %

  • nd

23.26 % 09.04 %

slide-26
SLIDE 26

26/38

Results 1:

pitch duration Germany China d

  • 32.41 %

43.73 %

  • d

49.57 % 72.56 % d

  • 04.51 %

22.53 %

  • d

24.13 % 57.70 % nd

  • 31.49 %

28.53 %

  • nd

23.26 % 09.04 %

slide-27
SLIDE 27

27/38

Results 1:

pitch duration Germany China d

  • 32.41 %

43.73 %

  • d

49.57 % 72.56 % d

  • 04.51 %

22.53 %

  • d

24.13 % 57.70 % nd

  • 31.49 %

28.53 %

  • nd

23.26 % 09.04 %

slide-28
SLIDE 28

28/38

Results 1:

pitch duration Germany China d

  • 32.41 %

43.73 %

  • d

49.57 % 72.56 % d

  • 04.51 %

22.53 %

  • d

24.13 % 57.70 % nd

  • 31.49 %

28.53 %

  • nd

23.26 % 09.04 %

slide-29
SLIDE 29

29/38

Results 2:

J(pab) C(pa) C(pb) 5 10 15 20 Corpus: Germ

any

| case: All phab | Pitch Analysis PI (Semitones) J(pab) C(pa) C(pb) 5 10 15 20

Corpus: China | case: All ph

ab | Pitch Analysis

PI (Semitones) Join Interval and Context Intervals Join Interval and Context Intervals

slide-30
SLIDE 30

30/38

Results 2:

J(pab) C(pa) C(pb) 5 10 15 20 Corpus: Germ

any

| case: All phab | Pitch Analysis PI (Semitones) J(pab) C(pa) C(pb) 5 10 15 20

Corpus: China | case: All ph

ab | Pitch Analysis

PI (Semitones) Join Interval and Context Intervals Join Interval and Context Intervals J(pab) C(pa) C(pb) 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Corpus: Germ

any

| case: All phab | Duration Analysis N IOI Join Interval and Context Intervals J(pab) C(pa) C(pb) 0.2 0.4 0.6 0.8 Corpus: China | case: All phab | Duration Analysis N IOI Join Interval and Context Intervals

slide-31
SLIDE 31

31/38

Results 2:

J(pab) C(pa) C(pb) 5 10 15 20 Corpus: Germ

any

| case: All phab | Pitch Analysis PI (Semitones) J(pab) C(pa) C(pb) 5 10 15 20

Corpus: China | case: All ph

ab | Pitch Analysis

PI (Semitones) Join Interval and Context Intervals Join Interval and Context Intervals J(pab) C(pa) C(pb) 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Corpus: Germ

any

| case: All phab | Duration Analysis N IOI Join Interval and Context Intervals J(pab) C(pa) C(pb) 0.2 0.4 0.6 0.8 Corpus: China | case: All phab | Duration Analysis N IOI Join Interval and Context Intervals

slide-32
SLIDE 32

32/38

Results 2:

J(pab) C(pa) C(pb) 5 10 15 20 Corpus: Germ

any

| case: All phab | Pitch Analysis PI (Semitones) J(pab) C(pa) C(pb) 5 10 15 20

Corpus: China | case: All ph

ab | Pitch Analysis

PI (Semitones) Join Interval and Context Intervals Join Interval and Context Intervals J(pab) C(pa) C(pb) 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Corpus: Germ

any

| case: All phab | Duration Analysis N IOI Join Interval and Context Intervals J(pab) C(pa) C(pb) 0.2 0.4 0.6 0.8 Corpus: China | case: All phab | Duration Analysis N IOI Join Interval and Context Intervals

slide-33
SLIDE 33

33/38

Results 3:

◮ group phrase pairs in respect to size(pha) ◮ pairwise t-test between m-j(pha,b) and m-c(pha,b) ◮ China PI: 19/26, NIOI: 26/26 ◮ Germany PI: 6/19, NIOI: 15/19

slide-34
SLIDE 34

34/38

Results 3:

6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 23 24 25 1.5 2 2.5 3 3.5 4 4.5 5 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 28 29 0.15 0.25 0.35 0.45 0.55 0.65

Length of pha

PI

NIOI

3 6 7 9 10 12 1.5 2 2.5 3 3.5 4 4.5 5 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 0.15 0.25 0.35 0.45 0.55 0.65

PI

Chinese Germanic

6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 23 24 25 1.5 2 2.5 3 3.5 4 4.5 5 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 28 29 0.15 0.25 0.35 0.45 0.55 0.65 3 6 7 9 10 12 1.5 2 2.5 3 3.5 4 4.5 5 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 0.15 0.25 0.35 0.45 0.55 0.65

NIOI Length of pha

slide-35
SLIDE 35

35/38

Results 3:

6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 23 24 25 1.5 2 2.5 3 3.5 4 4.5 5 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 28 29 0.15 0.25 0.35 0.45 0.55 0.65

Length of pha

PI

NIOI

3 6 7 9 10 12 1.5 2 2.5 3 3.5 4 4.5 5 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 0.15 0.25 0.35 0.45 0.55 0.65

PI

Chinese Germanic

6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 23 24 25 1.5 2 2.5 3 3.5 4 4.5 5 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 28 29 0.15 0.25 0.35 0.45 0.55 0.65 3 6 7 9 10 12 1.5 2 2.5 3 3.5 4 4.5 5 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 0.15 0.25 0.35 0.45 0.55 0.65

NIOI Length of pha

4.5 0.45 0.55 0.65 3 3.5 4 4.5 0.45

slide-36
SLIDE 36

36/38

Results 3:

6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 23 24 25 1.5 2 2.5 3 3.5 4 4.5 5 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 28 29 0.15 0.25 0.35 0.45 0.55 0.65

Length of pha

PI

NIOI

3 6 7 9 10 12 1.5 2 2.5 3 3.5 4 4.5 5 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 0.15 0.25 0.35 0.45 0.55 0.65

PI

Chinese Germanic

6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 23 24 25 1.5 2 2.5 3 3.5 4 4.5 5 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 28 29 0.15 0.25 0.35 0.45 0.55 0.65 3 6 7 9 10 12 1.5 2 2.5 3 3.5 4 4.5 5 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 0.15 0.25 0.35 0.45 0.55 0.65

NIOI Length of pha

4.5 0.45 0.55 0.65 3 3.5 4 4.5 0.45 2.5 0.25 2 2.5 0.25 0.35

slide-37
SLIDE 37

37/38

Conclusions

Under the scope of our study:

◮ discontinuities in pitch

  • are weak predictors of melodic phrase boundaries
  • can not be considered universal

◮ discontinuities in duration

  • are strong predictors of phrase boundaries
  • remain stable independent of cultural origin and size
slide-38
SLIDE 38

38/38

THANK YOU!