Musical Style Modification as an Optimization Problem
Frank ZalkowAudioLabs
Stephan BrandSAP SE
Benjamin GrafSAP SE
International Computer Music Conference, Utrecht, September 13, 2016
Content
1. Introduction and Overview
2. Data corpus
3. Local search within neighborhood
4. Objective functions
5. Results
© AudioLabs 2016
Slide 1 Frank Zalkow, Stephan Brand, Benjamin Graf
Introduction and OverviewMotivation
� Origin of project
� Musical style modification� Applications
© AudioLabs 2016
Slide 2 Frank Zalkow, Stephan Brand, Benjamin Graf
Introduction and OverviewMotivation
� Origin of project
� Musical style modification� Applications
© AudioLabs 2016
Slide 2 Frank Zalkow, Stephan Brand, Benjamin Graf
Introduction and OverviewMotivation
� Origin of project� Musical style modification
� Applications
© AudioLabs 2016
Slide 2 Frank Zalkow, Stephan Brand, Benjamin Graf
Introduction and OverviewMotivation
� Origin of project� Musical style modification� Applications
© AudioLabs 2016
Slide 2 Frank Zalkow, Stephan Brand, Benjamin Graf
Introduction and OverviewOverview of modification procedure
Musicinput
smallchanges. . .
Musicnew
Multipleobjectives
Scoringinput Scoringnew
Music withbetter scoringwill replace. . .
© AudioLabs 2016
Slide 3 Frank Zalkow, Stephan Brand, Benjamin Graf
Introduction and OverviewOverview of modification procedure
Musicinputsmall
changes. . .
Musicnew
Multipleobjectives
Scoringinput Scoringnew
Music withbetter scoringwill replace. . .
© AudioLabs 2016
Slide 3 Frank Zalkow, Stephan Brand, Benjamin Graf
Introduction and OverviewOverview of modification procedure
Musicinputsmall
changes. . .Musicnew
Multipleobjectives
Scoringinput Scoringnew
Music withbetter scoringwill replace. . .
© AudioLabs 2016
Slide 3 Frank Zalkow, Stephan Brand, Benjamin Graf
Introduction and OverviewOverview of modification procedure
Musicinputsmall
changes. . .Musicnew
Multipleobjectives
Scoringinput Scoringnew
Music withbetter scoringwill replace. . .
© AudioLabs 2016
Slide 3 Frank Zalkow, Stephan Brand, Benjamin Graf
Introduction and OverviewOverview of modification procedure
Musicinputsmall
changes. . .Musicnew
Multipleobjectives
Scoringinput Scoringnew
Music withbetter scoringwill replace. . .
© AudioLabs 2016
Slide 3 Frank Zalkow, Stephan Brand, Benjamin Graf
Introduction and OverviewOverview of modification procedure
Musicinputsmall
changes. . .Musicnew
Multipleobjectives
Scoringinput Scoringnew
Music withbetter scoringwill replace. . .
© AudioLabs 2016
Slide 3 Frank Zalkow, Stephan Brand, Benjamin Graf
Introduction and OverviewOverview of modification procedure
Musicinputsmall
changes. . .Musicnew
Multipleobjectives
Scoringinput Scoringnew
Music withbetter scoringwill replace. . .
© AudioLabs 2016
Slide 3 Frank Zalkow, Stephan Brand, Benjamin Graf
Content
1. Introduction and Overview
2. Data corpus
3. Local search within neighborhood
4. Objective functions
5. Results
© AudioLabs 2016
Slide 4 Frank Zalkow, Stephan Brand, Benjamin Graf
Data corpusPieces
Jaco Pastorius [Malone, 2002] Victor Wooten [Wooten, 2003]
© AudioLabs 2016
Slide 5 Frank Zalkow, Stephan Brand, Benjamin Graf
Data corpusPieces
Jaco Pastorius [Malone, 2002] Victor Wooten [Wooten, 2003]P. Metheny: Bright Size Life V. Wooten: A Show of HandsC. Parker: Donna Lee B. Fleck, V. Wooten and H. Levy: Blu-BopJ. Pastorius: Havona R. Noble: Cherokee (Indian Love Song)W. Shorter: Port Of Entry V. Wooten: Classical Thumb
J. Pastorius: Punk Jazz J. Lennon and P. McCartney: Norwegian Wood (This BirdHas Flown)J. Pastorius: Slang V. Wooten: Sex in a PanH. Mancini: The Days of Wine and Roses B. Fleck: Sinister MinisterJ. Pastorius: (Used To Be A) Cha Cha V. Wooten and B. Fleck: Stomping Grounds
© AudioLabs 2016
Slide 5 Frank Zalkow, Stephan Brand, Benjamin Graf
Data corpusPieces
Jaco Pastorius [Malone, 2002] Victor Wooten [Wooten, 2003]P. Metheny: Bright Size Life V. Wooten: A Show of HandsC. Parker: Donna Lee B. Fleck, V. Wooten and H. Levy: Blu-BopJ. Pastorius: Havona R. Noble: Cherokee (Indian Love Song)W. Shorter: Port Of Entry V. Wooten: Classical Thumb
J. Pastorius: Punk Jazz J. Lennon and P. McCartney: Norwegian Wood (This BirdHas Flown)J. Pastorius: Slang V. Wooten: Sex in a PanH. Mancini: The Days of Wine and Roses B. Fleck: Sinister MinisterJ. Pastorius: (Used To Be A) Cha Cha V. Wooten and B. Fleck: Stomping Grounds
2227.5 quarter length total 3642.75 quarter length total
© AudioLabs 2016
Slide 5 Frank Zalkow, Stephan Brand, Benjamin Graf
Data corpusPieces – Examples
Jaco Pastorius [Malone, 2002] Victor Wooten [Wooten, 2003]
Donna Lee
Z Z�Z Z3
Z Z ZZ Z Z
�A�
� ���� 44
Z = 218Z� ZF7Z
� Z� Z� � Z� Z� ZB� m7 Z
3
Z� Z Z�Z� Z Z Z Z
3 � ����ZB�7� Z
�Z�Z Z ZZ�
E� m7�Z Z Z� Z� Z�� Z�
Z�
3
ZD7� Z�ZZ Z�
Z� Z6 � ����
ZE�7
Z ZA�Z Z ZZ Z Z�
Z� Z� Z�Z� Z� ZA�� Z�� Z
� Z� ZZ� Z Z Z9 � �����D�� Z
�Z� Z�
D� m7Z
Z�Z � Z �Z� Z Z� Z
�3
3Z� � Z
Z Z Z� Z12 � ����ZF7
� Z Z ZB�7�� Z Z
Audio from J. Pastorius: Jaco Pastorius, Epic 1976.
© AudioLabs 2016
Slide 5 Frank Zalkow, Stephan Brand, Benjamin Graf
Data corpusPieces – Examples
Jaco Pastorius [Malone, 2002] Victor Wooten [Wooten, 2003]
Donna Lee Classical Thumb
Z Z�Z Z3
Z Z ZZ Z Z
�A�
� ���� 44
Z = 218Z� ZF7Z
� Z� Z� � Z� Z� ZB� m7 Z
3
Z� Z Z�Z� Z Z Z Z
3 � ����ZB�7� Z
�Z�Z Z ZZ�
E� m7�Z Z Z� Z� Z�� Z�
Z�
3
ZD7� Z�ZZ Z�
Z� Z6 � ����
ZE�7
Z ZA�Z Z ZZ Z Z�
Z� Z� Z�Z� Z� ZA�� Z�� Z
� Z� ZZ� Z Z Z9 � �����D�� Z
�Z� Z�
D� m7Z
Z�Z � Z �Z� Z Z� Z
�3
3Z� � Z
Z Z Z� Z12 � ����ZF7
� Z Z ZB�7�� Z Z
Z Z ZZ Z
Z Z ZZ ZZ
ZN.C.� �
44Z = 132
Z Z ZZ
Z Z ZZ Z
Z Z ZZZ Z3 � � Z
ZZ ZZ
Z Z Z Z ZZ Z Z
Z Z ZZ Z Z Z ZZ Z Z Z
Z5 � �Z Z Z Z
ZZ ZZ Z Z Z
Z Z ZZZ
Z Z ZZ Z Z
Z Z ZZ ZZ ZZZ Z
7 � �Z Z Z Z
Z Z ZZ ZZ Z
Audio from J. Pastorius: Jaco Pastorius, Epic 1976. Audio from V. Wooten: A Show of Hands, Compass 1996.
© AudioLabs 2016
Slide 5 Frank Zalkow, Stephan Brand, Benjamin Graf
Content
1. Introduction and Overview
2. Data corpus
3. Local search within neighborhood
4. Objective functions
5. Results
© AudioLabs 2016
Slide 6 Frank Zalkow, Stephan Brand, Benjamin Graf
Local search within neighborhood
Note events: N= (n1, n2, . . . , nI) where ni= (pi, di)
Chord events: C = (c1, c2, . . . , cJ) where cj= (sj , dj)
¾¾� �
� Changing the pitch pi
¾�¾� � �
� Changing the duration of two notes di1 and di2
�� � �
� Dividing ni into multiple notesÕ Õ�� �
� Joining ni and ni+1 into single noteh� �
© AudioLabs 2016
Slide 7 Frank Zalkow, Stephan Brand, Benjamin Graf
Local search within neighborhood
Note events: N= (n1, n2, . . . , nI) where ni= (pi, di)
Chord events: C = (c1, c2, . . . , cJ) where cj= (sj , dj)
¾¾� �
� Changing the pitch pi
¾�¾� � �
� Changing the duration of two notes di1 and di2
�� � �
� Dividing ni into multiple notesÕ Õ�� �
� Joining ni and ni+1 into single noteh� �
© AudioLabs 2016
Slide 7 Frank Zalkow, Stephan Brand, Benjamin Graf
Local search within neighborhood
Note events: N= (n1, n2, . . . , nI) where ni= (pi, di)
Chord events: C = (c1, c2, . . . , cJ) where cj= (sj , dj)
¾¾� �
� Changing the pitch pi
¾�¾� � �
� Changing the duration of two notes di1 and di2
�� � �
� Dividing ni into multiple notesÕ Õ�� �
� Joining ni and ni+1 into single noteh� �
© AudioLabs 2016
Slide 7 Frank Zalkow, Stephan Brand, Benjamin Graf
Local search within neighborhood
Note events: N= (n1, n2, . . . , nI) where ni= (pi, di)
Chord events: C = (c1, c2, . . . , cJ) where cj= (sj , dj)
¾¾� �
� Changing the pitch pi
¾�¾� � �
� Changing the duration of two notes di1 and di2
�� � �
� Dividing ni into multiple notesÕ Õ�� �
� Joining ni and ni+1 into single noteh� �
© AudioLabs 2016
Slide 7 Frank Zalkow, Stephan Brand, Benjamin Graf
Local search within neighborhood
Note events: N= (n1, n2, . . . , nI) where ni= (pi, di)
Chord events: C = (c1, c2, . . . , cJ) where cj= (sj , dj)
¾¾� �
� Changing the pitch pi
¾�¾� � �
� Changing the duration of two notes di1 and di2
�� � �
� Dividing ni into multiple notesÕ Õ�� �
� Joining ni and ni+1 into single noteh� �
© AudioLabs 2016
Slide 7 Frank Zalkow, Stephan Brand, Benjamin Graf
Local search within neighborhood
Note events: N= (n1, n2, . . . , nI) where ni= (pi, di)
Chord events: C = (c1, c2, . . . , cJ) where cj= (sj , dj)
¾¾� �
� Changing the pitch pi
¾�¾� � �
� Changing the duration of two notes di1 and di2
�� � �
� Dividing ni into multiple notesÕ Õ�� �
� Joining ni and ni+1 into single noteh� �
© AudioLabs 2016
Slide 7 Frank Zalkow, Stephan Brand, Benjamin Graf
Local search within neighborhood
Note events: N= (n1, n2, . . . , nI) where ni= (pi, di)
Chord events: C = (c1, c2, . . . , cJ) where cj= (sj , dj)
¾¾� �
� Changing the pitch pi
¾�¾� � �
� Changing the duration of two notes di1 and di2
�� � �
� Dividing ni into multiple notesÕ Õ�� �
� Joining ni and ni+1 into single noteh� �
© AudioLabs 2016
Slide 7 Frank Zalkow, Stephan Brand, Benjamin Graf
Content
1. Introduction and Overview
2. Data corpus
3. Local search within neighborhood
4. Objective functions
5. Results
© AudioLabs 2016
Slide 8 Frank Zalkow, Stephan Brand, Benjamin Graf
Objective functionsFeature classification
� Windowing (overlapping segments of 2–12 quarter lengths)
� 324 of music21’s feature extractors [Cuthbert et al., 2011] aswell as 86 customly designed ones
� Gradient Tree Boosting [Hastie et al., 2009], outputtingprobability
© AudioLabs 2016
Slide 9 Frank Zalkow, Stephan Brand, Benjamin Graf
Objective functionsFeature classification
� Windowing (overlapping segments of 2–12 quarter lengths)� 324 of music21’s feature extractors [Cuthbert et al., 2011] as
well as 86 customly designed ones
� Gradient Tree Boosting [Hastie et al., 2009], outputtingprobability
© AudioLabs 2016
Slide 9 Frank Zalkow, Stephan Brand, Benjamin Graf
Objective functionsFeature classification
� Windowing (overlapping segments of 2–12 quarter lengths)� 324 of music21’s feature extractors [Cuthbert et al., 2011] as
well as 86 customly designed ones� Gradient Tree Boosting [Hastie et al., 2009], outputting
probability
© AudioLabs 2016
Slide 9 Frank Zalkow, Stephan Brand, Benjamin Graf
Objective functionsFeature classificationExample of customly designed features:
{ �� � ����{ ��{
33
3
3
8va
3 3 3 ����
3 3 ��� ��
�(46)
44{{{{
� ���
��� �{��{��
fJaco =fLenb
mod1 wherea
b= fDur,
gcd(a, b) = 1
fLen = 4 and fDur = 1/3, so fJaco = 4/3mod1 = 1/3
© AudioLabs 2016
Slide 10 Frank Zalkow, Stephan Brand, Benjamin Graf
Objective functionsFeature classificationExample of customly designed features:
{ �� � ����{ ��{
33
3
3
8va
3 3 3 ����
3 3 ��� ��
�(46)
44{{{{
� ���
��� �{��{��
fJaco =fLenb
mod1 wherea
b= fDur,
gcd(a, b) = 1
fLen = 4 and fDur = 1/3, so fJaco = 4/3mod1 = 1/3
© AudioLabs 2016
Slide 10 Frank Zalkow, Stephan Brand, Benjamin Graf
Objective functionsFeature classificationExample of customly designed features:
{ �� � ����{ ��{
33
3
3
8va
3 3 3 ����
3 3 ��� ��
�(46)
44{{{{
� ���
��� �{��{��
fJaco =fLenb
mod1 wherea
b= fDur,
gcd(a, b) = 1
fLen = 4 and fDur = 1/3, so fJaco = 4/3mod1 = 1/3
© AudioLabs 2016
Slide 10 Frank Zalkow, Stephan Brand, Benjamin Graf
Objective functionsMarkov probability
� Separate Markov chains for durations and pitches
� Separate Markov chains per chord symbol type� Linear interpolation smoothing� Additive smoothing
© AudioLabs 2016
Slide 11 Frank Zalkow, Stephan Brand, Benjamin Graf
Objective functionsMarkov probability
� Separate Markov chains for durations and pitches� Separate Markov chains per chord symbol type
� Linear interpolation smoothing� Additive smoothing
© AudioLabs 2016
Slide 11 Frank Zalkow, Stephan Brand, Benjamin Graf
Objective functionsMarkov probability
� Separate Markov chains for durations and pitches� Separate Markov chains per chord symbol type� Linear interpolation smoothing
� Additive smoothing
© AudioLabs 2016
Slide 11 Frank Zalkow, Stephan Brand, Benjamin Graf
Objective functionsMarkov probability
� Separate Markov chains for durations and pitches� Separate Markov chains per chord symbol type� Linear interpolation smoothing� Additive smoothing
© AudioLabs 2016
Slide 11 Frank Zalkow, Stephan Brand, Benjamin Graf
Objective functionsRatio of example/counter-example Markovprobability
© AudioLabs 2016
Slide 12 Frank Zalkow, Stephan Brand, Benjamin Graf
Objective functionsRatio of example/counter-example Markovprobability
p1
394449545964697479
p2
394449545964697479
P̄(p1, p2)
0.000.010.020.030.040.050.060.07
© AudioLabs 2016
Slide 12 Frank Zalkow, Stephan Brand, Benjamin Graf
Objective functionsRatio of example/counter-example Markovprobability
p1
394449545964697479
p2
394449545964697479
P̄Pastorius(p1,p2)P̄Wooten(p1,p2)
05
10152025303540
© AudioLabs 2016
Slide 12 Frank Zalkow, Stephan Brand, Benjamin Graf
Objective functionsTime correlations for chord-repetitions
Z� Z Z Z
B�m7 Z Z Z¦ �� Z Z� ZZZ ZE�7ZZ� Z� Z
ZZ(29)
44����
�F7
ZA�
� ZZ Z Z
¦ �ZZ Z
Z� Z� Z � Z�Z
Z�
� Z� Z3
� Z
G�m7
� Z � Z
C�7
�Z� ZZ
E�7
� ���� 44(94) � Z�Z� Z
F�m7
0 2 4 6 8 10 12 14
time (ql)
45
50
55
60
65
70
pitc
h
0 1 2 3 4 5 6 7 8
time (ql)
45
50
55
60
65
70
pitc
h
0 2 4 6 8 10 12 14
time lag (ql)
−3
−2
−1
0
1
2
3
corr
elat
ion
×10−3
Z
Z
�Z
Z
�
�
Z
Z
�Z
Z
� Z
� Z
Z
Z
�Z
Z
� �
Z
� Z
Z
3
Z
Z
�
�
Z
Z
�
�
Z
Z
� �
Z¦ �ZZZ Z�
44����
� Z
44����
� Z ZZ�Z
� Z
Z
� Z
Z
��Z
¦ �
�Z
Z
Z
ZZ Z
© AudioLabs 2016
Slide 13 Frank Zalkow, Stephan Brand, Benjamin Graf
Objective functionsTime correlations for chord-repetitions
Z� Z Z Z
B�m7 Z Z Z¦ �� Z Z� ZZZ ZE�7ZZ� Z� Z
ZZ(29)
44����
�F7
ZA�
� ZZ Z Z
¦ �ZZ Z
Z� Z� Z � Z�Z
Z�
� Z� Z3
� Z
G�m7
� Z � Z
C�7
�Z� ZZ
E�7
� ���� 44(94) � Z�Z� Z
F�m7
0 2 4 6 8 10 12 14
time (ql)
45
50
55
60
65
70
pitc
h
0 1 2 3 4 5 6 7 8
time (ql)
45
50
55
60
65
70
pitc
h
0 2 4 6 8 10 12 14
time lag (ql)
−3
−2
−1
0
1
2
3
corr
elat
ion
×10−3
Z
Z
�Z
Z
�
�
Z
Z
�Z
Z
� Z
� Z
Z
Z
�Z
Z
� �
Z
� Z
Z
3
Z
Z
�
�
Z
Z
�
�
Z
Z
� �
Z¦ �ZZZ Z�
44����
� Z
44����
� Z ZZ�Z
� Z
Z
� Z
Z
��Z
¦ �
�Z
Z
Z
ZZ Z
© AudioLabs 2016
Slide 13 Frank Zalkow, Stephan Brand, Benjamin Graf
Objective functionsTime correlations for chord-repetitions
Z� Z Z Z
B�m7 Z Z Z¦ �� Z Z� ZZZ ZE�7ZZ� Z� Z
ZZ(29)
44����
�F7
ZA�
� ZZ Z Z
¦ �ZZ Z
Z� Z� Z � Z�Z
Z�
� Z� Z3
� Z
G�m7
� Z � Z
C�7
�Z� ZZ
E�7
� ���� 44(94) � Z�Z� Z
F�m7
0 2 4 6 8 10 12 14
time (ql)
45
50
55
60
65
70
pitc
h
0 1 2 3 4 5 6 7 8
time (ql)
45
50
55
60
65
70
pitc
h0 2 4 6 8 10 12 14
time lag (ql)
−3
−2
−1
0
1
2
3
corr
elat
ion
×10−3
Z
Z
�Z
Z
�
�
Z
Z
�Z
Z
� Z
� Z
Z
Z
�Z
Z
� �
Z
� Z
Z
3
Z
Z
�
�
Z
Z
�
�
Z
Z
� �
Z¦ �ZZZ Z�
44����
� Z
44����
� Z ZZ�Z
� Z
Z
� Z
Z
��Z
¦ �
�Z
Z
Z
ZZ Z
© AudioLabs 2016
Slide 13 Frank Zalkow, Stephan Brand, Benjamin Graf
Objective functionsTime correlations for chord-repetitions
Z� Z Z Z
B�m7 Z Z Z¦ �� Z Z� ZZZ ZE�7ZZ� Z� Z
ZZ(29)
44����
�F7
ZA�
� ZZ Z Z
¦ �ZZ Z
Z� Z� Z � Z�Z
Z�
� Z� Z3
� Z
G�m7
� Z � Z
C�7
�Z� ZZ
E�7
� ���� 44(94) � Z�Z� Z
F�m7
0 2 4 6 8 10 12 14
time (ql)
45
50
55
60
65
70
pitc
h
0 1 2 3 4 5 6 7 8
time (ql)
45
50
55
60
65
70
pitc
h0 2 4 6 8 10 12 14
time lag (ql)
−3
−2
−1
0
1
2
3
corr
elat
ion
×10−3
Z
Z
�Z
Z
�
�
Z
Z
�Z
Z
� Z
� Z
Z
Z
�Z
Z
� �
Z
� Z
Z
3
Z
Z
�
�
Z
Z
�
�
Z
Z
� �
Z¦ �ZZZ Z�
44����
� Z
44����
� Z ZZ�Z
� Z
Z
� Z
Z
��Z
¦ �
�Z
Z
Z
ZZ Z
© AudioLabs 2016
Slide 13 Frank Zalkow, Stephan Brand, Benjamin Graf
Content
1. Introduction and Overview
2. Data corpus
3. Local search within neighborhood
4. Objective functions
5. Results
© AudioLabs 2016
Slide 14 Frank Zalkow, Stephan Brand, Benjamin Graf
ResultsProblem of evaluation
© AudioLabs 2016
Slide 15 Frank Zalkow, Stephan Brand, Benjamin Graf
ResultsExample
Original iteration 0(0 accepted changes, target style Pastorius)
Z �F
Z �B�
3
Z ZN.C.Z �
F
�� � 43 ZZ Z �
F7Z
ZZZ Z �F
Z3
Z �F7
Z �6 � � �Z �
C7Z �C �
F�Z �Z Z �
C7
3
�Z �F12 � � �
B�� ZZZ
Z �F
Z �B�
3
Z ZN.C.Z �
F
�� � 43 ZZ Z �
F7Z
ZZZ Z �F
Z3
Z �F7
Z �6 � � �Z �
C7Z �C �
F�Z �Z Z �
C7
3
�Z �F12 � � �
B�� ZZZ
© AudioLabs 2016
Slide 16 Frank Zalkow, Stephan Brand, Benjamin Graf
ResultsExample
Original iteration 140(5 accepted changes, target style Pastorius)
Z �F
Z �B�
3
Z ZN.C.Z �
F
�� � 43 ZZ Z �
F7Z
ZZZ Z �F
Z3
Z �F7
Z �6 � � �Z �
C7Z �C �
F�Z �Z Z �
C7
3
�Z �F12 � � �
B�� ZZZ
ZZ �B�
Z3
�F
Z� Z ZF
�� � 43 ZN.C. Z Z Z �
F7
Z
Z Z ZZ Z �F
� Z3
�F7
ZZ �6 � � �
Z �C7Z �C �
Z �ZF�Z �
C7
3
��F
Z12 � � �
B�Z Z ZZ �
© AudioLabs 2016
Slide 16 Frank Zalkow, Stephan Brand, Benjamin Graf
ResultsExample
Original iteration 644(10 accepted changes, target style Pastorius)
Z �F
Z �B�
3
Z ZN.C.Z �
F
�� � 43 ZZ Z �
F7Z
ZZZ Z �F
Z3
Z �F7
Z �6 � � �Z �
C7Z �C �
F�Z �Z Z �
C7
3
�Z �F12 � � �
B�� ZZZ
ZZ� ��B�
3
�F
Z Z� Z ZF
�� � 43 ZN.C.
Z Z �F7
Z Z
Z Z Z ZF7Z �F
� Z3
Z� ZZ �Z6 � � �
Z �C7 Z�C �
ZFZ �Z �
C7
3
� �� �F12 � � �
B�Z � Z ZZZ
© AudioLabs 2016
Slide 16 Frank Zalkow, Stephan Brand, Benjamin Graf
ResultsExample
Original iteration 2058(14 accepted changes, target style Pastorius)
Z �F
Z �B�
3
Z ZN.C.Z �
F
�� � 43 ZZ Z �
F7Z
ZZZ Z �F
Z3
Z �F7
Z �6 � � �Z �
C7Z �C �
F�Z �Z Z �
C7
3
�Z �F12 � � �
B�� ZZZ
CB��
� �� �
�3
CF
��
� �F
��� � 43 �
N.C.
� � � �CF7�
��
� � � �F7� CF � �
3
�� �� C �6 � � C� �
C7 �CC
� F� C� CC7
3
�� CF12 � � C
B�� C � ���
© AudioLabs 2016
Slide 16 Frank Zalkow, Stephan Brand, Benjamin Graf
ResultsExample
Original iteration 1776(24 accepted changes, target style Wooten)
Z �F
Z �B�
3
Z ZN.C.Z �
F
�� � 43 ZZ Z �
F7Z
ZZZ Z �F
Z3
Z �F7
Z �6 � � �Z �
C7Z �C �
F�Z �Z Z �
C7
3
�Z �F12 � � �
B�� ZZZ
ZF
�Z �B�� Z
3
Z Z Z Z�� Z
Z �F�� � 43 ZN.C. Z Z �F
7Z
�F Z�C7 Z Z Z
3
Z Z �F7� Z Z Z�6 � � Z � Z
Z
�C ��
Z �F �Z �C7
Z�
3
Z � �
F
Z Z12 � � �
B�� � Z ZZ� �
© AudioLabs 2016
Slide 16 Frank Zalkow, Stephan Brand, Benjamin Graf
ResultsRecap
� Method for musical style modification
� Local optimization is great for this task� Hard to evaluate
© AudioLabs 2016
Slide 17 Frank Zalkow, Stephan Brand, Benjamin Graf
ResultsRecap
� Method for musical style modification� Local optimization is great for this task
� Hard to evaluate
© AudioLabs 2016
Slide 17 Frank Zalkow, Stephan Brand, Benjamin Graf
ResultsRecap
� Method for musical style modification� Local optimization is great for this task� Hard to evaluate
© AudioLabs 2016
Slide 17 Frank Zalkow, Stephan Brand, Benjamin Graf
ReferencesEthem Alpaydın. Introduction to Machine
Learning. Adaptive computation andmachine learning. MIT Press, Cambridgeand London, second edition, 2010.
Michael Scott Cuthbert, Christopher Ariza,and Lisa Friedland. Feature extractionand machine learning on symbolic musicusing the music21 toolkit. InProceedings of the 12th InternationalSymposium on Music InformationRetrieval, pages 387–392, 2011.
Trevor J. Hastie, Robert John Tibshirani,and Jerome H. Friedman. The elementsof statistical learning. Data mining,inference, and prediction. Springerseries in statistics. Springer, New York,2009.
Sean Luke. Essentials of Metaheuristics.Lulu, Raleigh, second edition, 2013.URL https://cs.gmu.edu/~sean/book/metaheuristics/.
Sean Malone. A Portrait of Jaco. The SoloCollection. Hal Leonard, Milwaukee,2002.
Victor Wooten. The Best of Victor Wooten.Transcribed by Victor Wooten. HalLeonard, Milwaukee, 2003.
Frank Zalkow. Automated musical styleanalysis. Computational exploration ofthe bass guitar play of Jaco Pastorius onsymbolic level. Master’s thesis,University of Music Karlsruhe, Germany,September 2015.
© AudioLabs 2016
Slide 18 Frank Zalkow, Stephan Brand, Benjamin Graf
https://cs.gmu.edu/~sean/book/metaheuristics/https://cs.gmu.edu/~sean/book/metaheuristics/
Introduction and OverviewData corpusLocal search within neighborhoodObjective functionsResults
