Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
1
Attempts to Reproduce a Pianist’sExpressive Timing with DirectorMusices Performance Rules
J. Sundberg, A. Friberg, and R. Bresin
Presented byBaharak Zali
ISE575/CSCI575/EE675, 2005
Literature/History/Background
n Research in music performancen Started in the 1970sn Co-operation with late Lars Frydenn First strategy: analysis-by-synthesis
n Simple Rules
n Director Musices (DM) performance grammarwas bornn Anders Friberg converted the rules into DM
Director Musices (DM)
n DM rules has two elements:n Context
n In what context the rules should be used
n Quantity parametern How great effects it should produce in the
performance
Director Musices (DM)n DM rules have different purposes:
n Grouping rulesn which tones belong togethern where the structure boundaries are
n Differentiation rulesn increase the differences between tone categories
including:n pitch classesn intervalsn note values
n Emphasis rulesn emphasize unexpected notes
2
Director Musices (DM)n Its purposes are relevant to speechn Capable of adding emotions to a performance
n By changing quantity parameter of variousperformance rules
n examples: no-expression, angry, sad, happy,tender, solemn, ... \Performance\examples\e-solemn.au
n Limitationsn System is developed based on limited number of
examplesn Can those rules be a general rule?
Methodn Idea: compare the real pianist’s performance
with the DM system performancesn Input: Score file in MUS format (a format that
allows the user to view and print the musicalscores), derived from MIDI
n Analysis1. Analysis of harmonic progression & musical
structure2. Chord symbols3. Phrase markers at structural boundaries
n phrase-start/end marker
Score used in the experiment
Fig. 1. Score used in the experiment. Numbers above staff refer to bar numbers, numbers withinparentheses below staff represent the levels of the phrase markers, numbers in the lower row showingthe start of phrases and numbers in the upper row marking the end of phrases.
Method
The piece was divided into 16 sections based the authors’interpretation of musical structure
3
Methodn Testing
n Top voicen Lower voices serve as enhancementn Analysis was limited to departures from nominal inter-onset-
interval (IOI)n Only rules affected IOI were tested (Table 1)
n Resultsn Evaluated by calculating the correlation between the
pianist’s departures and the ones DM generatedn Some tones were excluded
n different interpretations of ornaments
Rules used in experiments
Results
n Initial runn Effect of each rule affecting IOI was tested
separatelyn Agreement between pianist and DM displayed in
Fig. 2.n Single rules produced different effectsn Average of the correlation coefficients across
sections is lown Duration contrast, Faster uphill, Leap tone
duration, and Melodic charge hold low correlation
Fig. 2. Correlation between the pianist’s and the DM deviations from nominalduration obtained for the indicated sections by applying one rule at the time. Therightmost black bar indicates the average across sections of the correlations.
4
Results
n Second runn Combination of rules were testedn Combining the Phrase Arch rules at different levels
increased the correlation in several sections (Fig.3shows the comparison)
n Adding Harmonic Charge and Duration Contrast(Fig. 4)n Higher mean correlation means across sectionsn Reduced the negative correlation in several sections
n This version sounded acceptable
Fig. 3. Correlation between the pianist’s and the DM deviations from nominalduration obtained for the different sections by applying the indicated phrasingrule combinations. The rightmost black bar indicates the average across sectionsof the correlations.
Fig.4. Correlation between the pianist’s and the DM deviations from nominalduration obtained for the different sections by applying the rule combinationsindicated at the top. Numbers within parentheses show the k values. Therightmost bar indicates the average across sections of the correlations.
ResultsRule combination produced the highest correlation in each of the sections together
5
Resultsn Groupetto notes
n Definition? an 8th note followed by three triplet16th notes and another 8th note
n Pianistsn either shortened or lengthened some of the notes with
exact same duration (Fig. 5)n entire group’s length remained the same
n DM rulesn only one rule caused lengthening of one note in
groupettin a new rule were added to handle this
n Examples (Fig. 6)
Fig. 5. Pianist’s performance of gruppetti in the indicated bars and beats.The left and right panels show the deviation from nominal duration in ms and theright panel shows the inter-onset-intervals in ms. The heavy dashed lines in the leftpanel refers to the deviations produced by the new Note Triplet Contrast rule,applied with a quantity of k = 2.0.
ResultsEmphasizing
n Emphasis by Delayed Arrivaln Lengthening a note in order to emphasis
the next noten Used in music, singing, speechn Difficult to formulate
Fig. 6. Examples from the pianist’s performance of emphasis by delayedarrival (circled). The left panel refers to section A and the rightpanel to sections I and L. In the right panel the pitch names refer to section I.
6
Discussionn Correlation coefficient as a measure of
agreement between the pianist’s deviationfrom nominal and those generated by theDM.
n Limitations:n shows agreement with regard to the sign of the
overall deviations, not quantitiesn correlation highly sensitive to extreme valuesn more sensitive to the agreement for single notes
when the number of notes is small, compared towhen it is large
Discussion
n Some of the rules in isolation did notproduce positive correlation in anysection
n Better performance when rules are notapplied with a fixed constant quantitythroughout a piece
n Phrasing is truly a basic aspect of musicperformance
Questions
?