University of California Press
University of California Press
University of California Press
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact [email protected].
University of California Press is collaborating with JSTOR to digitize, preserve and extend access to Music
Perception: An Interdisciplinary Journal.
http://www.jstor.org
The Micro-andMacrostructural
Design
of ImprovisedMusic
JEFF PRESSING
La TrobeUniversity,Australia
Two short pieces of freely improvised music by the same performer were
recorded in microstructural detail by the use of a specially constructed
automatic transcription apparatus. The apparatus consists of a modified
DX7 synthesizer and 2650 microprocessor which interfaces with other
computers for data processing.
The resultant music is transcribed into a modified form of traditional
notation and subjected to both micro- and macrostructural analysis. Mi-
croanalysis includes the areas of timing (interonset and duration distri-
butions, displacement, chordal spreads, etc.), dynamics (key velocity,
quantization, chordal patterns, etc.), and legatoness (relative, absolute,
pedaling). Macroanalysis uses the full panoply of devices from tradi-
tional music theory (tonal procedures, rhythmic and motivic design,
pitch class sets, etc.). Correlations between microstructural parameters,
and with macrostructure, were found to be highly significant in Improvi-
sation A, which had a supplied external pulse, but largely absent in Im-
provisation B, which had no such pulse. Where pulse was present, rhyth-
mic design was found to be based largely on pulse subdivision and
shifting.
Some performance effects (e.g., chordal spreads) operated over a time
scale of 10 msec or less. Others (e.g., synchronization to an external
pulse) showed less resolution. Differences in the distribution patterns of
interonset times, durations, and legatoness suggest three independent un-
derlying temporal mechanisms that may sometimes link together in coor-
dination with macrostructure. Quantization ("categorical production")
of some variables (interonset times, key velocities) was clearly estab-
lished. The results were also interpreted in relation to an earlier model of
improvisation (Pressing, 1987).
Introduction
Requests for reprints may be sent to Jeff Pressing, Department of Music, La Trobe Uni-
versity, Bundoora, Victoria, Australia 3083.
133
Methods
APPARATUS
built-in or external controllers. Variations in intonation may also be studied. These poten-
tials will be explored in subsequent publications. With a minimum of adaptation the system
can also be made to function with a weighted 88-key master keyboard such as the Yamaha
KX88, removing most of the cited disadvantages.
EXPERIMENTAL PROTOCOL
DATA FORMAT
The standard data recorded comprise key velocity on a scale of 1 to 127, and key onset,
key offset, and sustain pedal times all to the nearest millisecond. The microprocessor feeds
these data to a PDP 11/43 and a numerical printout is obtained, with automatic calculation
of note durations, likely note duration categories, legatoness, and other parameters. A pre-
cise bar graph in what is effectively proportional notation, with width indicating note dura-
tion, midpoint indicating pitch height (key number), and thickness indicating key velocity, is
also output. The MIDI data representation (as mentioned above, less accurate- particularly
for chords) can be used to drive a Personal Composer software program that automatically
produces an approximate rendition of the piece in traditional notation, which can be useful
as an intermediate step in the production of final precise notation. This last technique was
not used for the pieces described in this article.
Key velocity (KV) data require special comment. All DX7s put out a reduced range of key
velocities relative to other synthesizers. The maximum KV value output by this DX7 was
120, with values above 117 extremely rare regardless of the force applied to the keys. This is
typical. Typical apparently also is a certain fine-grained artifactuality, whereby, for exam-
ple, on this DX7, even KV values are extremely rare in the ranges 94-98 and 112-114, and
odd ones equally rare in the ranges 85-87 and 103-107. This irregularityled to a perceived
need for calibration of the KV data relative to an independent measure of key transit time.
Such a calibration was therefore performed, where, for a large number of random key
strokes, KV data were recorded in conjunction with independent measurement of key transit
time, using an oscilloscope. The results show that KV is roughly linear with measured key
transition time, as expected. An estimate of the range of error in the measured KV values was
obtained as follows. For individual representative KV values across the full dynamic range,
the corresponding weighted distribution of key transit times (KTT) that could have gener-
ated the chosen KV value was noted. The probability that various KV values would be pro-
duced from this distribution of KTTs was then read off the calibration graph. This gives the
error distribution around the input KV value. When this is done, it is found that the error in
KV values is ±2-3 units at at least 80% confidence level throughout the full dynamic range
employed. The error is actually less than this at lower dynamic values (^90).
A final point concerns the relationship of key velocity to sound level. This will vary with
the synthesizer voice, since velocity sensitivity is fully programmable in the DX7 FM al-
gorithms. A calibration of a voice of the type used in this study (PIANO 1) was therefore
performed, using a Bruël and Kjaer Impulse Precision Sound Level Meter, Type 2209, accu-
rate to ±0.5 dB. The results of this calibration are displayed in Figure 1 . Obviously the abso-
lute sound level is dependent on the amplifier gain.
IMPROVISATIONA
Fig. 1. Decibels versus key velocity (KV) for the voice PIANO 1.
timesarefoundin columnsthreeandfourrespectively.Succeedingcolumns
providenote durationin seconds,durationin percentageof the beat,closest
musicallynotateddurationwithin a frameworkof dupleor triplesubdivi-
sion, errorin this notatedapproximation(firstin secondsand then in beat
%), note displacementfrom the beat (firstin secondsand then in beat %),
interonsettime, absolute legato (overlapin seconds between the current
andpreviousnotes), and a relativelegatonessfactor(definedbelow).
Figure3 shows an excerpt from the computer-producedproportional
notationrepresentationof the piece.Pitchheightis on the verticalaxis, time
on the horizontal.Each note is representedby a black rectangularregion
whose length gives its duration,thicknessthe key velocity, and midpoint
the key number.
Macrostructural Analysis
The analysishere is based on a reductionof the complete data into a
slightlymodifiedform of traditionalnotation, as shown in Figure4. This
was done by hand.The legendexplainsthe use of specialsymbols;primär-
t
i <
i «ti
V)
I
I
1
CO
Uh
C/5
§
q
r^
o
oo
q
•2'Ì
x s
B °
S§
"C O
^.i
1^
Is
1§
2 a
1
1.2
g
l§
5-.»
1 O <u
S .S
Second,therearedistinctiverecurringmotivesandintervallicsequences.
Most noticeableperhapsare threerelatedorderedsets, occurringas indi-
catedin Figure5, foundprimarilyin the bass.Thesemotives(orderedsets)
consistof a minor6th, 5th, or tritone,followedby a perfect4th. In interval
numbernotation, this class of three motives could thus be representedas
[(678),5]. In action representationthis generalizedmotive can be seen to
correspondto the left handfingering521, with slightlyvariablestretchbe-
tweenthe forefingerandlittlefinger.Thisraisesthe possibilitythat the mo-
tive class is stored to a significantdegree in kinestheticrepresentation,
ratherthan aural.
J = 6° [4msec]
, ,
ft ^T^? <^TTj>. («">7\
==
^ 93 tr; F Ïo3 ,05
DX7 102 ^_
<- ^
104
99
104
concurrent 34 93 102 93
tncnordUl4
trichord
014
thoughts
2 handdescendinggesture fourths
10477 82 94 109 H ^ ^ J- 95 I
»' io
102 -Jq ~^^^^™i 106 I 103
*83 ^9-
w ^-97^ 93 6 llOO
102 T T trichord
contrary chordscorrect? grabbedat
motion last moment
Jffi*> ^ fiL»-| i s ?a
[10 msec]
[26 msec] I 3 1
B
i\\ - 28-29
S 9m 38: containedin r.h. chord
m
-_ 38-39: transposedup a 4th, retrograded,in bass
40-41
43-44: 8va, permuted,in bass
48-49: 8va, retrograded
63-65: top voice, transposed3 octaves +
semitone,intervallicallyinverted
C
fry , 13: transposedup 8va + 4th
JA * \ m= 18
üZZ 41 : transposed, intervallically inverted, rearranged,
IT*" in treble
43: transposed up 8va + M2nd, in bass
47-48: transposed up 8va + M2nd, in bass
51, 63, 65: transposed up 2 or 3 octaves + mi3rd,
intervallically inverted, in treble
Fig. 5. Motives and ordered sets in Improvisation A.
Microstructural Analysis
Timing
Timing analysis here includes the following: interonsettimes, note and
chorddisplacement,durations,and chordalspread.Interonsettime (IO)is
the time between successivenote attacks. If we label attack and release
timesof the ith note playeda{and r,respectively,then
di = ri-ai (2)
TABLE 1
Displacement Categories
Displacement
Center Strength Interpretation
Dynamics
QuantizationFigure10 shows a plot of key velocityincidence.KVvalues
rangefrom 1 to 112, with a meanof 91.8 andstandarddeviation18.9. The
Legatoness
One advantageof microstructuralrecordingis the possibilityof examining
degreesof legatoness.In the followingdiscussion,this is definedin termsof
the overlapof adjacentnotes, where adjacentmeans adjacentin orderof
attacktime. Thus therehas not been, as would be expectedin traditional
musicanalysis,a separationinto parts,whose legatonesswould then each
be consideredseparately.We are lookingthereforeat "global"ratherthan
"line"legatoness.Becauseof the thinnessof textureand the natureof the
figurationsin the two improvisationshere, this differenceis not a major
one. But it is also not always possible to reliablydistinguishindependent
partsin the transcriptionof improvisedmusic.In any case, the questionof
globallegatonessis a legitimateone on its own.
AbsoluteLegatonessabsolute (key) legatonessL, of any pair of adjacent
notes may be definedas theiroverlaptime, viz.
TABLE 2
Interpretations of Peaks
Peak Location Interpretation
Correlations
Here we discuss correlationsbetween microstructuralparametersand
withmacrostructure. To facilitatethiscomparison,Figure14 shows a num-
berof microvariablesplottedagainsttime, scaledto allow inclusionon the
samepage.Somenoteswith extremeabsolutelegatovalueshavebeenomit-
ted for clarity.
An additionalvariableis also occasionallydiscussed.This is notational
shift,which refersto smalldeviationsfrom the attacktimes of transcribed
notation,wheresimple(experientially,"felt")beatsubdivisionsareusedby
the performer.This reducesto note displacementfor on-beats.
The discussionwill not treatthe entirepiece in full detail, but will look
selectivelyat significantdetailswithinchosenphrasesto highlightthe kinds
of correlationsfoundin this improvisation.
table 3
Absolute Legatoness of the Motives
Beat
30 31 32 33
Event Partitioning
Ina recentpaper(Pressing,1987), I presentedwhat is apparentlythe first
cognitivemodel of the improvisationprocess.The model is basicallysyn-
thetic(as opposedto analytic)in that it describeshow improvisationsmay
be generated.It can only be analyticto the extent that sufficientperform-
anceinformationis availableto allow the unambiguousparsingof the mu-
sic into groupsof musicalevents,underspecifiedcriteria.This provedpos-
sible for ImprovisationA and gives support for the applicabilityof the
theory.
The model considersthat any improvisationI can be brokenup into a
seriesof nonoverlappingeventclustersE, suchthat
I=[EUE2...EN], (6)
whereeachE,is an integrallyconceivedmotorieandmusicalunitconsisting
table 4
Partitioning of Improvisation A
EventCluster BeatLocation EventClusterClass
10.780 sec, with mean 5.065 sec. The choices for the boundariesbetween
the Ei and Kfwere madeon musical(phrasestructures,motives,etc.), mo-
torie (inferredfingering,hand position, etc.) and cognitive(knowledgeof
the performer'sstyle,recordedcommentsin score)grounds.
IMPROVISATIONB
Macrostructural Analysis
A full transcriptionof this piece is seen in Figure15, wherepedal mark-
ings are given along the horizontalaxis. Sinceno click trackwas used to
"drive"note production,such a transcriptioninto traditionalnotation is
onlypossibleif somepulseis chosenas fundamental.In the examinationof
note durationand IO time quantization(see below) no dominantfunda-
mentalpulseemerged,nor did the performerclaimto be usingone. There-
foreJ = 60 was chosenfor transcriptionconvenience,whichis at leastcon-
sistentwith an observedweak IO peak at about 1 sec and allows ready
comparisonwith ImprovisationA. However, it must be rememberedin
lookingat Fig. 15 that no implicationsof tempo, pulse or metershouldbe
drawn.Therefore,except where rapid passageworkdemandsshort note
values,durationsareroundedfor clarityby as muchas 150 msec,to avoid
complicatedandunreasonableliteralness.In all cases,however,local musi-
cal structure(orderof attacks,presenceof legatoness,etc.) is preservedin
the notation.
ImprovisationB in manyways displaysfarless extensivemacrostructure
thanImprovisationA. Therearefew recurrentmotives,andpitchmaterials
show less homogeneity.The usage of sets shows no simple centralunity.
The lack of referenceto a definedpulse has presumablycontributedto
make this overall a "freer"improvisation.Nevertheless,the following
pointsdo hold.
In generalterms,the piece is slow moving, and dominatedby sustained
textures,pedal points, and conjunctmotion. Phrasestructureearlyin the
piece(beats6-19, 20-35 and36-50) is balancedandfairlylong.Thisthen
breaksinto fasternote values and shorterasymmetricalphrases,particu-
larly in the passage of beats 69-81. The final passage (beats 81 -end)
shows a simplerdurationalstructureand a returnto somethinglike the
originaltexture.
(J = 60)
6 7 I , 9 « h. i3 . is | [-12msec]
[-18 msec]
29 1-30msec) 3 33 35 37
À*L\ ]J^ JJ
^-^
'
99
^ "
IJ l>^r~li .8899
IM1 f
, r"'
-^ iL^Jj 106
^
msecn [9 msec]
^
[9:llmH 4i 43 45 47 [17 msec] & msecl
h [
j m iœ
W^|S^
'
i;j:r
^^.^"'l 91
^ ^[jj^
68 ^ '
[17 msec]
Microstructural Analysis
Timing
InteronsetTimesIO timesin this piece,as mightbe expected,show weaker
quantizationeffectsthanin ImprovisationA. Figure16 shows that only the
Dynamics
QuantizationFigure18 shows that thereare clearquantizationeffectsop-
eratinghere. Keyvelocitiesvariedfrom 1 to 111 and theirmeanwas 87.0,
with standarddeviation19.5. Largepeaks are found, in the two-unitwin-
dow plot, centeredat 91, 99, and 106. These correspond
closely with the
majorpeaksfoundfor ImprovisationA, at 93, 100, and 108 KVunits,and
Legatoness
Absolute Legato Significant detail for absolute (key) legato was only found
in the region below 200 msec, which is plotted in Figure 19. There are clear
peaks centered at 45 and 105 msec. The 45-msec peak was also found in
Improvisation A, but a weak peak around 105 msec there was not judged
significant. This suggests weakly that absolute legato quantization can be
based on pulse and also operate independently of it. Further research is
clearly necessary, but apparently legatoness forms a third type of quantiza-
ble time framework, after IO time and duration.
Relative Legato Relative (key) legato Xis plotted in Figure 20. This does not
include any pedaling effects. There are three weak peaks: at 0.08, 0.19, and
0.54. A comparison with Improvisation A suggests that the peaks at 0.08
and 0.54 correspond to peaks there, shifted slightly (0.04-0.05) to the
right. As there, they may be ascribed to a tenuto and 50% legato process.
The absence of the peak at 1 .00, as in A, is clearly due to the use of the pedal.
The peak at 0.19 does not have an obvious explanation.
Pedaling The presence of extensive pedaling should emphasize that the sec-
tions on duration and legatoness just discussed have referredexclusively to
the actions of the hand. The major effects of pedaling may be gleaned di-
Correlations
In contrastwith ImprovisationA, only veryfew micro-microor micro-
macro correlationswere found in ImprovisationB. IO times, durations,
and absoluteand relativelegatonessshowedno clearmicrostructuralrela-
tions to any of the passage-specificmotivesor processesdescribedas mac-
rostructure.No correlationwas foundbetweenchordalspreadandchordal
dynamicspread.Dynamiccorrelationswere found to be absent from the
M6 motivein its variousforms,andfromthe long conjunctstructurallines.
The only correlationsfound were those associatedwith the dynamicsof a
few stronglyidentifiablemotives of short duration:the chromaticgrace
note figureand the closingM7 motives.Specifically,the threenotes of the
chromaticallyascendinggracenote figureand their final main note were
alwaysplayedwith crescendo.The four M7 motives showed a consistent
Event Partitioning
Discussion
References
1 . The apparatus was built by the La Trobe University Department of Psychology work-
shop, with software written by Russell MacDonnell and Glen Lawrence of the same depart-
ment. I am also grateful to the La Trobe University School of Humanities for a grant for a
purchase of the DX7 synthesizer and to Greg Troup for experimental assistance and useful
discussions.