Transcript Document

A preliminary
computational model
of immanent accent
salience in tonal music
SysMus
Richard Parncutt1, Erica Bisesi1, & Anders Friberg2
1University of Graz, Austria
2KTH Stockholm, Sweden
Research object
(example)
Chopin Prélude in A major
performed by Claudio Arrau
Bisesi, Parncutt, Friberg
Method: Performance rendering
Aim:
Understand performance
- not replace the performer
Approach:
Empirical quantitative science
1. Develop a theory
2. Implement it as an algorithm
3. Test its predictions
kulturserver-nrw.de
Too many variables!  Isolate them
1. Separate composer (score) from performer
2. Consider only timing and dynamics (piano)
Bisesi, Parncutt, Friberg
What motivates expressive
piano performance?
Aim: What is the performer trying to achieve?
Means: On that basis, what do we expect?
1. Aim: Participate in a cultural tradition
Means: Imitation of well-known performance patterns
2. Aim: Speak to the audience
Means: Pseudo-random variation (speech without phonemes)
3. Aim: Communicate gesturally with the audience
Means: Sound patterns based on physical gestures (kinematic)
4. Aim: Communicate musical structure to the listener
Means: Emphasis of structurally important events
4
Bisesi, Parncutt, Friberg
Musical structure
Global:
Intermediate:
Local:
form
phrasing
accents
A pianist can emphasize:
The start or end of a new section
The start or end of a phrase
An important note or chord
Tillmann, Bigand, and Madurell (1998)
5
Bisesi, Parncutt, Friberg
A taxonomy of accent
Bisesi, Parncutt, Friberg
A two-stage model of
performance rendering
1. Analyse structure and estimate
salience of immanent accents
2. Adjust timing and dynamics in
the vicinity of accents
7
Bisesi, Parncutt, Friberg
1. Immanent accents: Subjective salience estimates
Structurally important events in Chopin’s Prélude in A major
Erica E. Bisesi
Bisesi, Parncutt, Friberg
2. Performed accents at immanent
accents: Subjective salience estimates
Chopin Prelude op. 28 n. 13
5
Subjective evaluation of recorded
performances of 16 eminent pianists
4
(means and standard deviations)
melodic
harmonic
metric
grouping
5
4
3
3
2
2
accent salience
1
1
0
0
0
3
6
9
12 15 18 21 24 27 30 33 36 39 42 45 48 51 54 57 60 63 66 69 72 75 78 81 84 87 90 93 96
9
Models of timing and dynamics
near accents
Bisesi, Parncutt, Friberg
Sample predictions
to evaluate subjectively or compare with recordings
Bisesi, Parncutt, Friberg
A preliminary computational
model of immanent accent
salience in tonal music
•
•
•
•
Grouping
Metrical
Melodic
Harmonic
Bisesi, Parncutt, Friberg
Grouping accent salience
•
•
Start and ends of phrases
Hierarchically structured
Estimate accent salience
Simple model: hierarchical depth
Complex : sum of salience at each level
Bisesi, Parncutt, Friberg
Procedure
• Divide piece into 2 or 3 sections
• Divide each section into 2 or 3 (etc.)
• Follow composer’s markings
Metrical accent salience
Metrical level
Time Level Level Level Level
0
2
3
signa1
(beat)
ture
4/4
1/8
1/4
2/4
4/4
2/2
1/4
1/2
2/2
4/2
4/2
1/4
1/2
2/2
4/2
2/4
1/8
1/4
2/4
4/4
3/4
1/8
1/4
3/4
6/4
3/8
1/16 1/8
3/8
6/8
6/8
1/8
3/8
6/8 12/8
9/8
1/8
3/8
9/8 18/8
Bisesi, Parncutt, Friberg
Melodic accent salience
Assumed to depend on:
• distance from mean pitch
• size of preceding leap
• whether peak or valley
Procedure
Calculate (local) mean pitch
Assign two values, S1 and S2, to each note
S1 = |interval from mean in semitones|
(if pitch is below mean, multiply S1 by 0.7)
S2 = |preceding interval in semitones|
(if interval is falling, multiply S2 by 0.7)
Melodic salience = S1 * S2
Bisesi, Parncutt, Friberg
Harmonic accent salience
Calculated accent saliences
Not including phrasing (grouping accents)
Bisesi, Parncutt, Friberg
Calculated accent saliences
Not including phrasing (grouping accents)
Bisesi, Parncutt, Friberg
Next…
Computer interface
• Representation of score with accents
• Pop-up boxes for timing/dynamic functions
Psychological testing
• Listener ratings of artificial performances
Stylistic issues
• Performer styles
• Intended emotions
• Shifts within and between pieces
Combine with other approaches?
• Cultural (arbitrary learned patterns)
• Aleatoric (speech-like)
• Gestural (kinematic)
Bisesi, Parncutt, Friberg
A preliminary computational model of
immanent accent salience in tonal music
Richard Parncutt1, Erica Bisesi1, & Anders Friberg2
1University of Graz, Austria
2KTH Stockholm, Sweden
SysMus
An approach to performance rendering based on
• music analysis: accent
• music psychology: communication of structure
1. Analyse score for immanent accents
(grouping, metrical, melodic, harmonic)
2. Estimate the perceptual salience of each
3. Manipulate timing and dynamics near each