Music Perception © 1994 BY THE REGENTSOF THE

Summer1994, Vol. 11, No. 4, 409-464 UNIVERSITYOF CALIFORNIA

A PerceptualModel of PulseSalienceand MetricalAccent

in MusicalRhythms

RICHARD PARNCUTT
McGill University

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

In Experiment 1, six cyclically repeating interonset interval patterns (1,
2:1, 2:1:1, 3:2:1, 3:1:2, and 2:1:1:2) were each presented at six differ-
ent note rates (very slow to very fast). Each trial began at a random
point in the rhythmic cycle. Listeners were asked to tap along with the
underlying beat or pulse. The number of times a given pulse (period,
phase) was selected was taken as a measure of its perceptual salience.
Responses gravitated toward a moderate pulse period of about 700 ms.
At faster tempi, taps coincided more often with events followed by
longer interonset intervals. In Experiment 2, listeners heard the same set
of rhythmic patterns, plus a single sound in a different timbre, and were
asked whether the extra sound fell on or off the beat. The position of the
downbeat was found to be quite ambiguous.
A quantitative model was developed from the following assumptions.
The phenomenal accent of an event depends on the interonset interval
that follows it, saturating for interonset intervals greater than about 1 s.
The salience of a pulse sensation depends on the number of events match-
ing a hypothetical isochronous template, and on the period of the
template- pulse sensations are most salient in the vicinity of roughly
100 events per minute (moderate tempo). The metrical accent of an
event depends on the saliences of pulse sensations including that event.
Calculated pulse saliences and metrical accents according to the
model agree well with experimental results (r > 0.85). The model may
be extended to cover perceived meter, perceptible subdivisions of a beat,
categorical perception, expressive timing, temporal precision and dis-
crimination, and primacy/recency effects. The sensation of pulse may be
the essential factor distinguishing musical rhythm from nonrhythm.

Introduction

Imagine that you are walking down a quiet city alley and enter a jazz
club. The door opens and suddenly you hear the music. In just a second or
two you have a strong impression of the "feel" of the music- in particu-

Dedicated to Prof.Dr.-Ing. E. Terhardton the occasion of his 60th birthday.

Requests for reprints may be sent to Richard Parncutt, Faculty of Music, McGill Univer-
sity, 555 Sherbrooke West, Montreal, Quebec, Canada H3A 1E3.

409
410 Richard Parncutt

lar, the way it "swings"- its "beat" or "pulse." If instead you had been
confronted with some Ghanaian percussion music, or a disco number, or
Ravel's Bolero, or a movement of a Bach Motet driving inexorably for-
ward in 3/4 time toward a glorious climax, you would have received an
entirely different qualitative impression of the music's beat. But the experi-
ence would have been similar in two respects: the strength of the impres-
sion of beat (swing, pulse), and the remarkably short time required to
perceive it.
Yeston (1976) referred to an isochronous sequence of similar-sounding

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

events as a rhythmic stratum, or level of motion (also called time level by
Jones & Boltz, 1989, level of pulsation by Palmer & Krumhansl, 1990,
and rhythmic level by Rosenthal, 1992). In Yeston's theory, a rhythmic
level may be defined by note onsets, accents of all kinds (dynamic, me-
lodic, harmonic, timbrai, textural, phrasal), and repetition of a temporal
pattern. Integral to the theory is the idea that a single rhythmic sequence
can evoke several rhythmic levels at the same time. A listener may focus
attention on any one of these, or switch attention from one to another at
will (Jones, Boltz, & Kidd, 1982; Jones, Kidd, & Wetzel, 1981).
Yeston divided the various rhythmic levels evoked by a rhythmic se-
quence into one, fastest level or pulse level, and slower (higher) interpreta-
tive levels. Interpretativelevels whose periodicity is slower than that of the
barlines are referred to as hypermeters by Rothstein (1989); straddling the
conventional borderline between rhythm and form, they play an impor-
tant role in most tonal-rhythmic music. In the present study, the term pulse
sensation is used to describe all rhythmic levels spontaneously evoked in
the mind of the listener. The term pulse sensation is intended as a blanket
term for "beat," "swing," "rhythmic level," and so on.
The widespread use of rubato (especially in Western classical music)
indicates that a musical beat need not be exactly isochronous, or equally
spaced in time, to be strongly felt (Shaffer, 1981). In general, however,
stricter timing tends to induce a stronger feeling of pulse, and the strength
of the beat evoked by a sequence of sounds tends to decrease as the
sequence deviates from isochrony (cf. Ehrlich, 1958). Thus, African percus-
sion music (as described by Agawu, 1987; Jones, 1959; Koetting, 1986;
Rowlands, 1991) generally evokes a stronger feeling of pulse than does
European classical music.
Rhythmically organized auditory patterns are easier to encode, recall,
and reproduce than similarly organized visual patterns (Garner & Gott-
wald, 1968; Glenberg, Mann, Alman, Forman, & Procise, 1989; Handel
& Buffardi, 1968). A feasible explanation is that presentation times are
encoded more accurately in the auditory case (Glenberg & Swanson,
1986). This may in turn be due to the confinement of the sensation of beat
or pulse to the auditory modality (as demonstrated, e.g., by Grant &
 Pulse Salience and Metrical Accent 411

LeCroy, 1986). Perhaps auditory rhythms can be perceived by encoding

their constituent events relative to an underlying musical beat or pulse, but
visual "rhythms" cannot. This view would be consistent with observations
of Glenberg and Jona (1991) and Schab and Crowder (1989) that the
advantage of auditory over visual rhythms diminishes for tasks involving
relatively long durations, because the perceptual salience of pulse sensa-
tions diminishes as their period increases from about 600 to 2000 ms
(Fraisse, 1982).
Schulze (1978) investigated the sensitivity of listeners to various kinds

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

of timing deviation in isochronous sequences. His results suggested that
the listeners synchronized an internal time-keeper with the sequence, de-
tecting irregularities by monitoring discrepancies between expected and
actual event times. Wing and Kristofferson (1973a, 1973b) applied the
concept of a timekeeper to a model of periodic motor timing. Shaffer
(1981) and Shaffer, Clarke, and Todd (1985) analyzed the timing of perfor-
mances of piano music and concluded that timing was constrained by a
central timekeeper capable of adjusting its rate in response to expressive
features of the music (rubato). Povel (1984) hypothesized that listeners
match an isochronic temporal grid (called a template in the present model)
to a rhythmic sequence, choosing the grid in such a way as to directly
match a maximal number of tones and to account for the other tones with
the greatest ease (principle of economy). Povel and Essens (1985) and
Drake and Gérard (1989) assumed that listeners generate an internal clock
or time base while listening to a rhythmic pattern.
The terms internal clock and pulse sensation refer to the same phenome-
non, but differ in emphasis. "Internal clock" alludes to an underlying
neurophysiological mechanism; "pulse sensation," to the experience of the
listener. Pulse sensations may be experienced during rhythmic perception
(listening), rhythmic action (performance), or both. In this sense, the idea
of pulse sensation is consistent with an ecological approach to perception
(Gibson, 1979), in which perception and action are regarded as inextrica-
bly linked and sensations are regarded as byproducts of the interaction
between an organism and its environment.

SERIALVERSUS PERIODIC GROUPING

Sound events in speech may be structured either serially (concatenated

in associative chains), hierarchically, or both (Martin, 1972). In music,
this distinction may be described in terms of figurai coding or grouping on
the one hand, and meter on the other (Bamberger,1978, 1982; Cooper &
Meyer, 1960; Deutsch, 1982; Jones, 1976; Lerdahl & Jackendoff, 1983;
Rosenthal, 1989; Upitis, 1987; Vuori, 1991). A problem with these terms
is that meter itself may be regarded as a kind of grouping: It groups events
412 Richard Parncutt

into equivalence classes, such as all nth beats of a bar, by analogy to pitch
classes or chroma (Benjamin, 1984). Jones (1987b) avoided the semantic
ambiguity of the term grouping by distinguishing between horizontal (se-
rial) and vertical (metrical) components of rhythmic production. In the
present study, the terms serial and periodic temporal grouping are used,
and meter is regarded as a form of periodic grouping.
The relative importance of periodic and serial grouping depends on the
listener. For infants, serial grouping is generally more important than
periodic grouping (Bamberger, 1980; Drake, Dowling, & Palmer, 1991),

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

whereas the rhythmic perception and performance of adult musicians re-
flects an organization in which periodic grouping predominates (Smith,
1983). But the effect of age and training on perceptual grouping does not
influence all aspects of rhythmic activity. Povel (1981), for example, failed
to find significant differences between the results of musicians and
nonmusicians in a rhythmic imitation task.
Serial grouping depends primarily on the serial proximity in time, pitch,
and timbre of temporally adjacent events (Bregman & Pinker, 1978;
Fraisse, 1956; Martin, 1972; Miller & Heise, 1950; van Noorden, 1975).
Gaps between serial groups tend to be the largest gaps available in a given
sequence (Garner, 1974; Handel, 1974; Vos, 1977). According to Lerdahl
and Jackendoff (1983), serial grouping in music includes "motives, themes,
phrases, periods, theme-groups, sections, and the piece itself" (p.12); and
"only contiguous sequences can constitute a group" (p. 355).
Periodic grouping usually depends on the relative timing and perceptual
properties of nonadjacent events (Martin, 1972). Periodic grouping may
be divided into two stages, here called pulse sensation and perceived meter.
The second stage involves the simultaneous perception of different pulses
(or multiple temporal periodicities: Palmer & Krumhansl, 1990). The
result is a regular alternation of strong and weak beats, corresponding to
the generally accepted definition of meter (e.g., Lerdahl & Jackendoff,
1983, p. 12). For example, if the tempi of two pulses stand in the ratio 1 :2,
then a binary pattern of strong and weak beats results.
Cooper and Meyer (1960) claimed that serial groups arrange them-
selves around accented events, implying that serial grouping depends on
periodic grouping. Similarly, Benjamin (1984) suggested that both accents
and serial grouping influence metrical organization, implying that serial
grouping influences periodic grouping. Lerdahl and Jackendoff (1983)
advanced the simpler (and hence more testable) hypothesis that serial and
periodic grouping are independent and must therefore be kept separate:
"groups do not receive metrical accent, and beats do not possess any
inherent grouping" (p. 26). Similarly, Povel (1984) distinguished between
the organization of elements (serial grouping) and the organization of
intervals (periodic grouping), remarking that "though presumably occur-
 Pulse Salience and Metrical Accent 413

ringin parallel,they may to a greateror lesserextent be incompatible"(p.

332).
The assumption of independenceof serial and periodic grouping is
fundamentalto the present study. Serial and periodic grouping are as-
sumed to depend in separateand distinctways on the timing (interonset
intervals[IOIs])and otherperceptualproperties- pitch, loudness,timbre,
and articulation- of sound events. Accordingly,the model presentedbe-
low emulates periodic grouping and metrical organization in simple
rhythmswithout takingserialgroupinginto account.

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

The remainderof this paper may be divided into two parts. The first
part describes an experimentalinvestigationof periodic grouping in a
specificset of rhythms.Experiment1 concernsthe salienceof pulse sensa-
tions; Experiment2, the strengthof metricalaccents. The experiments
provide data to enable the developmentand testing of a quantitative
modelof pulse salienceand metricalaccent,which will be describedin the
secondpart.

Experiment 1: Pulse Salience

Differentrhythmiclevelsor pulsesensationsarerarelyheardto be equal

in significance(Krebs,1987). Normally,one level is heardas primaryand
acts as a frameof referencefor the (conscious)perceptionof other levels.
Jones and Boltz (1989) called this the referenttime level Accordingto
LerdahlandJackendoff(1983, p. 21), "Thelistenertendsto focus primar-
ily on one (or two) intermediatelevel(s) in which beats pass by at a
moderaterate . . . Adaptingthe Renaissanceterm,we call such a level the
tactus" The present experimentinvestigatedthe tactus of some simple
rhythms.
The perceptual(and hence musical) significance,strength,or promi-
nenceof a pulse sensationmay be calledits salience.The tactusis then the
pulse sensationwith the highestsalience.Musicalexperiencesuggeststhat
the salience of pulse sensations depends on both tempo and rhythmic
pattern.For example,some rhythmicpatterns"swing"more than others;
and piecesmay lose theirswing,or simplyfeel wrong,if playedtoo slow or
too fast. Variations in tempo can also affect subjectiveaccentuation.
Michon (1974) found that relativetemporalstresses(durationalaccents)
in a performanceof the Vexationsby Erik Satie varied as a function of
performancetempo, implyinga dependencyof subjectivetemporalstruc-
ture on tempo. Clarke (1982) related these variations to variations in
groupingstructure,observingthat the musictendedto be segmentedinto
fewer serialgroupsat fastertempl.
The present experimentinvestigatedeffects of rhythmicpattern and
414 Richard Parncutt

tempo on periodic grouping by measuring the salience of the various pulse

sensations evoked by a range of rhythms. Listenerstapped along with each
rhythm at equally spaced intervals, in a similar way that a jazz musician or
listener taps along to a complex jazz rhythm. Pulse salience was then
estimated by counting the relative number of times each pulse-train re-
sponse was selected.
Related experiments were performed by Handel and Lawson (1983)
and Povel and Essens (1985, Experiment 3). Handel and Lawson investi-
gated the perception of polyrhythms consisting of two concurrent pulse

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

trains, distinguishing the two stimuli by playing them at different pitches.
Povel and Essens measured the apparent simplicity of concurrently pre-
sented rhythms and pulse trains, again at different pitches. The present
study differs in that the rhythm presented in each trial consisted entirely of
identical sounds. The only other sound heard during the experimental
trials was that of the computer's space key being struck or depressed. This
sound is unlikely to have influenced the results, because by the time it was
heard, the listener had already decided on a pulse response, and was in the
process of tapping it out. The model presented later in this study is simi-
larly restricted to rhythms composed of physically identical sound events.

METHOD

Listeners

Twenty-two students (from the University of Stockholm) and researchers (from the
Department of Speech Communication and Music Acoustics, Royal Institute of Technol-
ogy, Stockholm) took part. Their musical experience (defined as the number of years
regularly practicing and/or performing a musical instrument, including voice) covered a
wide range: minimum 0 years, maximum 30 years, mean 12 years.

Apparatus

Stimuli were drum strokes on various percussion instruments, digitally recorded on a

commercially available drum machine (Roland TR-505 Rhythm Composer). Timing of
rhythms was controlled via Musical Instrument Digital Interface (MIDI) by a Le_ Lisp
program on a Macintosh II computer. Rhythms were amplified and reproduced over a
small loudspeaker in a sound-isolated room.

Stimuli

As shown in Figure 1, six different rhythmic patterns were crossed with six different
tempi, making 36 trials in all. The first four of the rhythmic patterns (pulse, waltz, march,
swing) were metrically relatively simple and unambiguous. The last two (skip, cross) were
more ambiguous: They may be perceived in either 3/2 or 6/4 meter (that is, as either 3
groups of 2 beats, or 2 groups of 3) depending on tempo (cf. Handel & Oshinsky, 1981).
One of the aims of the present experiment was to investigate this ambiguity.
 Pulse Salience and Metrical Accent 415

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

Fig. 1. Reference table for stimuli used in the experiments. Each pattern is given a name for
ease of identification. The boxes define rhythmic cycles or measures. The name cross refers
to a cross-rhythm of 1/2-notes and 3/4-notes.

Tempi were specified by the number of notes or events per unit time, as this allowed the
full range of musical tempi to be covered more satisfactorily for each pattern than specify-
ing the number of beats per unit time. The experimenter adjusted the note rates to cover as
wide a range as possible within the musical and physical constraints of the experiment, by
subjective evaluation of each of the six patterns when played at different rates. The note
rates that were adopted were logarithmically equally spaced and ranged from 50 to 400
notes per minute, or 0.83 to 6.7 notes per second. The corresponding IOIs of the nominal
1/4-note beats in each pattern are shown in Figure 1.
Early events in a rhythmic sequence are more likely to be interpreted as downbeats than
are later events (cf. Longuet-Higgins ÔCLee, 1982). In the present experiment, a starting
point effect was undesirable, as the aim was to investigate the perceptual properties of
cyclically repeating rhythms as a function only of rhythmic pattern and note rate. The
starting-point effect may be softened by starting a sequence at a subliminal intensity and
gradually increasing the level (Vos, 1973). But this does not eliminate the effect, as it is still
generally possible to define the first audible event. Alternatively, the sequence may be
started at a very fast tempo and then made to decelerate toward the intended tempo (Royer
& Garner, 1970). This was not a practical alternative in the present study, where tempo
was itself an important independent variable. Here, the starting-point effect was avoided
simply by starting rhythms at random temporal points in the cycle. For example, pattern
(b) waltz was presented in either of two forms: 121212 or 212121 (where 1
= 1/4-note, 2 = 1/2-note in Figure 1). The probability of a pattern starting on a certain
note was made proportional to the interonset interval that would otherwise have preceded
it. So waltz was twice as likely to start with a 1/4-note than with a 1/2-note.
The attention of the listener was maintained by randomly selecting the timbre (instru-
ment) to be used in each trial from sounds available on the drum machine (labeled low
conga, high conga, timbale, low cowbell, high cowbell, hand clap, ride cymbal, bass drum,
snare drum, low torn, mid torn, high torn, rim shot, closed high-hat, open high-hat).
Timbre remained constant throughout each trial. The loudness of the various instruments
had been equalized by ear before the experiment.
416 RichardParncutt

Procedure

In each trial, listeners were asked to press or tap the space bar of a computer keyboard
in time with the underlying beat of the rhythm. They were permitted to tap in any way they
wished; most used the index finger, middle finger, or both together, of their dominant
hand. A tap was deemed to have coincided with a particular 1/4-note (as no tated in Figure
1) if it fell in a temporal category of width one 1/4-note, centered on the onset of the note.
In other words, temporal category boundaries for taps were set midway between 1/4-note
onsets. This procedure is consistent with the categorical perception of rhythmic patterns as
described by Clarke (1987) and Schulze (1989).
In each trial, tap times were recorded until such time as four consecutive tap times were
equally spaced, indicating a consistent pulse response. At this point, the stimulus stopped,

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

and the response was recorded. No feedback was given. Trials were presented in a different
random order for each listener. The experiment was preceded by a short practice session.
Raw data were the periods (time between taps) and phases (relative to the nominal start
of the rhythm) of the tapped pulses, in 1/4-note units. The terms period and phase corre-
spond to the terms unit and location of Povel and Essens (1985). Other possible terms for
period are grid interval (Povel, 1984), subdivision (Povel & Essens, 1985), metrical unit
(Essens & Povel, 1985), beat interval (Summers, Hawkins, & Mayers, 1986), and cardi-
nality (Krebs, 1987).

RESULTS AND DISCUSSION

Figure 2 shows the number of times each pulse response was selected
(plain text). The results are compared with predictions according to the
model (italics). The comparison between the results and predictions will
be discussed later.

Fig. 2. Results of Experiment 1 (pulse salience) for all 161 selected pulses. In the table on
the right side of each panel, plain text denotes numbers of times each pulse was selected;
italics denote corresponding calculated values according to Equation 7 with free parameter
values t = 550 ms, i = 1.6, /* = 760 ms, a = 0.23, and ; = 2.0. Each column corresponds
to a particular combination of rhythmic pattern and note rate, or to one of the 36 trials in
the experiment.
 Pulse Salience and Metrical Accent 417

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

Fig. 2. (Continued) Calculations for each trial are normalized (multiplied by a constant) so
as to add up to 22, the number of listeners. The numbers 1-6 (top right) are rate numbers
(see Figure 1). The periods and phases shown on the far left are expressed in nominal 1/4-
note beats. Phases are shown only if necessary for the unambiguous specification of pulse
responses; otherwise, a dash (-) is marked.
418 Richard Parncutt

Each response was specified by four parameters: the rhythmic pattern

and note rate of the stimulus, and the period and phase of the response.
Each of the 36 trials of the experiment allowed for a number of different
possible pulse responses (periods and phases). In all, 161 different pulse
responses were collected. The number of different pulse responses as a
function of pattern was 17, 24, 23, 29, 37, and 31; as a function of rate,
22, 28, 23, 24, 29, and 35. These summary data tentatively suggest that
(1) the main pulse sensation or tactus evoked by each combination of
rhythmic pattern and note rate was quite ambiguous and (2) the ambiguity

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

of the tactus increased as the rhythmic patterns became more complex. In
addition, (3) ambiguity showed a small but inconsistent tendency to in-
crease as note rate increased.
Figure 2a shows results for the six different presentation rates of pattern
(a) pulse. At the slowest rate (no. 1), all 22 listeners tapped in time with
every sound event (that is, with period = one 1/4-note beat). At the second
rate, 18 listeners tapped in time with every sound event (period = 1 beat)
and 4 with every second event (period = 2 beats). The higher the tempo of
the isochronous sequence, the more pulse sensations it evoked- that is,
the higher was its rhythmic ambiguity. This is consistent with the observa-
tion (e.g., Handel & Lawson, 1983) that when a rhythm is played more
slowly, responses will tend to be faster relative to the rhythm, and vice-
versa. In other words, pulse responses tend to gravitate toward a moderate
tempo.
The perceptual grouping of isochronous sound events at moderate to
fast tempi is called subjective rhythmization. In general, groups of four
occur more often than groups of three (see references under Enhancement
of pulse salience below). Figure 2a suggests that this effect is quite general
and independent of tempo.
In the waltz rhythm (Figure 2b) at a very slow tempo (rate 1), 16 of the
22 listeners tapped with a period of one 1/4-note. Of the remaining six
listeners, three tapped on the "implied" or "missing" events between the
notes (period 3, phase 1), two tapped in time with the 1/4 notes (period 3,
phase 2), and one tapped a cross-rhythm (period 2). No one tapped in time
with the 1/2-note onsets - the music-theoretical downbeat (period 3,
phase 0). At faster rates, the proportion of listeners tapping the music-
theoretical downbeat increased until, at the highest rates, some tapped
with every other downbeat (period 6, phase 0). Responses with periods 5
and 7 were clearly errors; in the interests of objectivity, however, they were
not removed from the data.
Vos, Collard, and Leeuwenberg (1981) played recorded performances
of Bach Preludes and asked listeners to tap at equal time intervals as they
listened. The location of the downbeat was found to be quite ambiguous:
 Pulse Salience and Metrical Accent 419

Some listeners tapped in synchrony with the notated barlines, while others
tapped at the same rate but out of phase with the barlines. A similar effect
was found in the present results for the march rhythm (Figure 2c), for
responses with period 4 at slow to moderate note rates. For rates 1-4,
results were low for both phase 0 and phase 2. For rates 5 and 6, however,
phase 0 (the music-theoretical downbeat) predominated over phase 2. A
similar effect of tempo was observed in the swing (Figure 2d) and skip
(Figure 2e) rhythms at period 6, phases 0 and 3 (or 4). In general, the
results indicate that long events are more important than short events in

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

fast rhythms, but not necessarily in slow ones.
In the last three rhythmic patterns, a tendency was observed for some
listeners to "drag," or tap a little later than expected (see results with
period 6, phase 1). This observation is consistent with the idea of a motor
delay between intended and actual responses, as incorporated into the
theory of self-paced periodic tapping of Wing (1973) (described by
Vorberg & Hambuch, 1978). In other tapping experiments, however, lis-
teners have been found to anticipate events (Fraisse, 1980, 1987). In the
present experiment, it appears that some listeners simply had insufficient
time to catch up with fast rhythms before their response was collected and
the stimulus ceased.
= 792
Figure 3 shows the distribution of the periods of all 22 x 36
pulse trains selected in Experiment 1. The histogram was determined by
calculating logarithms to the base 10 of the periods in milliseconds and
then allocating them to categories of width 0.1. The distribution has a
mean of -0.149 (corresponding to a period of 710 ms) and a standard
deviation of 0.224 (corresponding to the range 420-1190 ms). The bell-
shaped curve corresponds approximately to the existence region of pulse
sensation (formulated in Eq. 6 of the model presented below). The low
results at -0.05 on the horizontal axis (for logarithms between -0.1 and
0.0, or periods between 790 and 1000 ms) appears to be an artifact of the
experimental procedure, due to the relatively low number of possible
responses in this range (remembering that the same set of note rates was
used for all rhythmic patterns).

Experiment 2: Metrical Accent

Sequences of identical events spaced at different IOIs may produce

subjective accents. Subjective accents have been investigated for isolated
pairs of tones (Buytendijk & Meesters, 1942) and for sequences of tones
containing two alternating time intervals (Povel & Okkerman, 1981). The
aim of the present experiment was to investigate subjective accentuation in
420 Richard Parncutt

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

Fig. 3. Further results of Experiment 1 (pulse salience). Crosses: distribution of the
periods of all 22 x 36 = 792 pulse responses. Vertical axis: number of pulse responses
falling in categories of width 0.1 (in log10 of pulse period in seconds). Curve: The
corresponding normal distribution.

simple rhythmic patterns. The same set of six patterns was examined as in
Experiment 1.
In music theory, different beats in a measure are assigned different
accent strengths (see, e.g., Lerdahl & Jackendoff, 1983). In a 4/4 measure,
for example, the first 1/4-note beat is the strongest, the third beat is the
second strongest, and the second and fourth beats are the weakest. In a 6/4
measure, the second-strongest beat is the fourth of the six 1/4-note beats
(phase = 3); but in a 3/2 measure, the second-strongest beat is the fifth
1/4-note beat.
Palmerand Krumhansl (1990) measuredthe relative salience of the differ-
ent beats in standard musical meters by the following method. They first
presented a low-pitched pulse train with a period in the range 1.7-4.8 s and
asked their listeners to imagine each pulse as the first of two, three, four, or
six isochronous beats. After four pulses, a higher-pitched probe tone was
sounded between the pulses, corresponding to one of the imagined beats.
The listeners then rated how well the probe tone fit the imagined metrical
context. The results were consistent with music-theoreticalnotions of metri-
cal accent and confirmed that their listeners had quite detailed and stable
knowledge of typical Western meters.
The present experiment differed from that of Palmer and Krumhansl in
that it aimed to measure perceptual properties of real-time (rather than
 Pulse Salience and Metrical Accent 421

imagined) temporal sequences. Listeners in the present experiment were

not asked to imagine the subdivision of the bar; instead, the subdivision
was suggested directly by the temporal sequence they heard.

METHOD

Listeners

Twenty-onepeopletook part,all of whom had participatedin the previousexperiment.

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

Stimuli

Thesamesix rhythmicpatternswerepresentedas in Experiment1, butonlyone noterate

was used: 150 notesper minute.As before,eachtrialbeganat a randompoint in the cycle.
After8-10 notes of the rhythm,a singletargetnote was playedin a differenttimbre(bass
drum).Thetargetcoincidedwithone of thebeatsin thecycle,butnot necessarilywithone of
the notes. The rhythmthen continuedwithout interruptionuntil the listenerresponded.
The experimenttested each individual1/4-note beat in the cycle of each of the six
rhythmicpatternsin Figure1. The cyclelengthsof the six patternswere 1, 3, 4, 6, 6, and 6
beats, so the total numberof trialswas 26. Instrumentaltimbreswere selectedrandomly
fromthe same set as before,with the exceptionof the bass drum,whichwas reservedfor
the targetbeat.

Procedure

Listenerswere askedto indicatewhetherthe targetwas on or off the beat,usinga four-

point ratingscale. Labelsfor the scale were providedfor all listenersin both musicaland
nonmusicallanguage.The musicallabelswere:0, verysyncopated;1, quitesyncopated;2,
on a weakbeat;3, on a strongbeat.Thenonmusicallabelswere:0, off the beat(sure);1, off
the beat (not sure);2, on the beat (not sure);3, on the beat (sure).Listenerswere free to
choosewhicheverlabelstheypreferred.Theexperimentbeganwith a shortpracticesession.

RESULTSAND DISCUSSION

In subjective rhythmization, an isochronous sequence evokes a pulse

sensation whose period is longer than that of the stimulus. It follows that
some events in isochronous sequences may sound off the beat. This expec-
tation was borne out in the single result for pattern (a) pulse, shown in
Figure 4a, which was significantly lower than the maximum of the re-
sponse scale.
The results for the other rhythmic patterns (Figures 4b to 4f) agreed
with music-theoretical expectations, with some minor but interesting ex-
ceptions. Results for pattern (c) march peaked at the third 1/4-note beat
instead of the first, contradicting the idea that events followed by longer
IOIs have stronger accents. A similar effect had been suggested in Experi-
ment 1 (see Figure 2c): At rate 4 (174 events per minute, which is close to
the rate of 150 events per minute used in Experiment 2) and period 4
422 Richard Parncutt

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

Fig. 4. Results of Experiment 2 (metricalaccent) for rhythmicpatterns (a) pulse, (b) waltz, (c)
march, (d) swing, (e) skip, and (f) cross. The horizontal axis of each panel is marked with the
note values of each pattern as shown in Figure 1. Points: Mean responses (21 data each).
Bars: 95% confidence intervals of means. Squares: Calculations according to Equation 8,
with free parameters t = 180 ms, i = 1.7, /i = 660 ms, cr = 0.14, and ; = 1.9. The 26
calculations have been linearly adjusted to have the same mean and standard deviation as the
26 mean responses.

(corresponding to the cycle length), four listeners tapped on the first beat
(phase 0), whereas seven tapped on the third (phase 2). It appears in this
case that the IOI preceding the event had a slightly greater effect on its
subjective accent than did the IOI following the event. According to Povel
and Essens (1985, p. 415), the initial and final events in a cluster of three
events have subjective accent; here, the accent on the initial event of the
group was stronger than the accent on the final event.
Results for swing (Figure 4d) at the first and fourth 1/4-note beats were
not significantly different from each other, that is, the downbeat was
about equally likely to fall on either beat. This is consistent with the results
of Experiment 1 for swing at rate 4, period 6, phases 0 and 3 (Figure 2d).
The downbeat of skip was similarly ambiguous, falling on either the first
 Pulse Salience and Metrical Accent 423

or the fifth beat (cf. Figure 2e, rate 4, period 6, phases 0 and 4); and the
downbeat of cross was perceived with about equal likelihood on beats 1,
3, or 5 (cf. Figure 2f, rate 4, period 6, phases 0, 2, 4).
These results of Experiment 2 only apply for a note rate of 150 notes/
min. The results of Experiment 1 suggest that differences between event
saliences would have been greater, that is, the meter of the rhythms would
have been less ambiguous, at faster templ.

Model

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

The need for a quantitative, computational model of the perception of
meter was recognized by Simon (1968). Since then, models of meter have
been developed by Longuet-Higgins and Steedman (1971), Longuet-
Higgins (1976), Longuet-Higgins and Lee (1982, 1984), Povel and Essens
(1985), Lee (1991), and Rosenthal (1992). The models of Longuet-
Higgins and Lee simulated real-time listening by beginning at the start of a
rhythm and testing hypothetical metrical interpretations at each new
event. By contrast, the present model, like the models of Povel and Essens
and of Rosenthal, looks at a whole rhythmic sequence at once, testing all
possible metrical interpretations (clocks, rhythmic levels) for goodness of
fit. Tentative extensions of the present model to real-time processing are
proposed below under the heading Primacy,recency, and the psychological
present.
A central assumption of the present model is the inherent ambiguity of
the underlying pulse (tactus) and meter of a rhythm. The model does not
output a single solution, but instead considers many possible pulse and
meter sensations, estimating the relative importance or salience of each. In
this respect it is similar to the models of Povel and Essens (1985), Lee
(1991), and Rosenthal (1992). Metric ambiguity is particularly important
in the case of hypermeter- meter that is perceived but not notated and
whose period exceeds that of a notated measure (Rothstein, 1989).
The present model is based on an earlier version (Parncutt, 1985),
which in turn was based on an exposition of the theory of pulse sensation
(Parncutt, 1987). The original model had been tested by comparing its
predictions with music-theoretical analyses and intuitions. Modifications
to the present model were inspired by a combination of experimental
results and theoretical considerations. Appropriate mathematical forms
were chosen according to the subjective requirements of theoretical clarity
and parsimony and the objective requirements of computational sim-
plicity and efficiency. Changes to the model were adopted only if they
enhanced correlation coefficients between predictions and experimental
424 Richard Parncutt

data. The same procedure also enabled values of free parameters to be

estimated.
The model is primarily intended to account for the results of the two
experiments just described and to link them together, predicting metrical
accent from pulse saliences. The model also suggests explanations for
various further rhythmic phenomena. The perception of meter may be
assumed to involve the simultaneous perception of concurrent pulses,
and the salience of a perceived meter may be modeled by adding calcu-
lated pulse saliences. Other phenomena that may be clarified by the

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

present approach include limitations on the number of allowable subdivi-
sions of a beat, the categorical perception of time in rhythms, the percep-
tibility of small temporal deviations in isochronous sequences, expressive
timing in musical performance, and the temporal development of pulse
salience and meter. Finally, the theory suggests a concise definition of
musical rhythm.
A schematic representation of the model is shown in Figure 5. The input
to the model is a cyclically repeating pattern of IOIs in milliseconds. The
sound events making up the rhythm are assumed to be physically identical.
The model begins by assigning to each event in a rhythmic cycle a phe-
nomenal accent, calculated according to a saturation function relating
phenomenal accent to IOI (see Eq. 3, Figure 6 below). Next, the percep-
tion of pulse is simulated by a pattern-matching routine (see Eq. 4), and
the tempo dependence of pulse salience is accounted for in terms of an
existence region of pulse sensation (see Eq. 6). This leads to a prediction of
pulse salience (see Eq. 7), defined as a measure of the probability of
occurrence of a given tactus or underlying beat. Calculated pulse saliences
are compared with the results of Experiment 1. The results of Experiment
2 are predicted by estimating the metrical accent at each (actual or im-
plied) rhythmic event, by summing the calculated saliences of all pulse
sensations converging on that temporal category (see Eq. 8). Furtherappli-
cations of the model are discussed at the end of the paper.

INPUT

assumption: The relationship between physical and perceived time is

essential linear.
On the basis of a thorough literature survey, Allan (1979) found that
perceived time is in general almost proportional to physical time, suggest-
ing that no distinction need be drawn between physical and perceived time
in a model of rhythm perception. Time is measured in the present model
either absolutely (in milliseconds) or relative to a basic time unit b in
milliseconds. The values for b used in the above experiments were the 1/4-
 Pulse Salience and Metrical Accent 425

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

Fig. 5. Schematic representation of the model.

note beats given in Figure 1. Absolute time in milliseconds is denoted t;

relative time in basic time units, T.
The model is presently restricted to cyclically repeating rhythms. The
length of a cycle is denoted either c (in ms) or C (in basic time units). For
example, pattern pulse in Figure 1 has C = 1, waltz has C = 3, march has
C = 4, and the other patterns all have C = 6;c additionally depends on the
number of events per minute in a given trial. The cycle is assigned an
arbitrary starting point where t = T = 0. In Figure 1, this point corre-
sponds to the first note in each pattern.
The basic time unit is assumed to be perceived as a temporal category.
One rhythmic cycle contains C such categories. In the trivial case of an
isochronous sequence, all temporal categories are represented by sound
events. In nonisochronous rhythms, some categories are "empty." The
model begins by assigning to each "full" time category an IOI indicating
the number of categories (basic time units) between that category and the
next "full" category. Intervening "empty" categories are assigned IOI = 0.
426 Richard Parncutt

For example, the waltz rhythm of Figure 1 had IOI (0) = 2, IOI (1) = 0,
and IOI (2) = 1. The IOIs add up to the cycle length:

2 IOI (T) = C (1)

T=0,C-l

PHENOMENAL ACCENT

assumption: Phenomenal accents may be induced by changes in IOI,

loudness, timbre, and pitch and by combinations thereof.

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

According to Lerdahl and Jackendoff (1983), rhythmic accents are of
three kinds: phenomenal, structural, and metrical. Phenomenal accents
may be evoked by changes in IOI (Povel & Essens, 1985; Povel &
Okkerman, 1981), loudness (dynamic accent), timbre (e.g., Wessel, 1979),
melodic contour (Thomassen, 1982; Woodrow, 1911), and implied har-
mony (Smith &cCuddy, 1989). Jones (1987a) referredto the superposition
of different patterns of accentuation as joint accent structure and to the
combination of different kinds of accent at a given temporal position as
accent coupling. In general, the phenomenal accent Ao at temporal cate-
gory T may include contributions from durational accent Ad, dynamic
(loudness) accent Ab timbrai accent Av and pitch accent Ap (where Ap has
melodic and harmonic components), as well as possible interactions be-
tween these parameters:

AO{T)= Ad(T) + A,(T) + At(T) + Ap(T) + interactions (2)

Of the various contributions to phenomenal accent, durational accent
appears to have the greatest impact on metrical organization (cf. Palmer,
1989). Several pieces of evidence support this claim. An increase of 2 dB in
the level of a tone in an isochronous sequence is sufficient to produce a
dynamic accent (Thomassen, 1982), but an increase of about 4 dB is
needed to produce a dynamic accent strong enough to balance an dura-
tional accent (Povel & Okkerman, 1981). Young children can reproduce
temporal but not dynamic structures of musical rhythms (Gérard &
Drake, 1990), suggesting that the temporal structure of simple rhythms is
more perceptually salient than their dynamic structure. Finally, the vari-
ability of IOIs in tapped patterns is considerably less than the variability of
intensities (Brown, 1911) - suggesting that phenomenal accent is more
sensitive to changes in IOI than to changes in loudness.
Phenomenal accent may also be influenced by articulation. The articula-
tion of an event may be defined as the ratio of the onset-to-offset interval
(OOI) to the IOI after the event: articulation = OOI/IOI. According to
this definition, articulation approaches a value of one in legato and about
 Pulse Salience and Metrical Accent ATI

one-half or less in staccato. Musical experience suggests that articulatory

accents produced by playing particular notes legato rather than staccato
(e.g., in the articulation of two-note phrases) are generally weaker than
accents due to variations in IOI (cf. Sloboda, 1983, and studies cited below
under expressive timing). In other words, the auditory system tends to
respond to short-duration events in rhythmic sequences as if they lasted
until the onset of succeeding events.
In the present version of the model, phenomenal accent depends only on
IOI. In this regard, the present model is similar to other published models

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

of rhythm and meter (Longuet-Higgins & Lee, 1982; Povel & Essens,
1985; Rosenthal, 1992). The stimuli in the above experiments were also
limited to variations in IOI; loudness, timbre, pitch, and physical event
duration were held constant in each trial.

DURATIONALACCENT,ECHOIC STORAGE,AND MINIMUM

DISCRIMINABLEIOI

assumption: The subjective accentuation of an event increases with the

IOI that follows it.
In general, the greater the IOI after an event, the greater is (1) its
phenomenal accent or perceptual salience; (2) its contribution to the for-
mation of pulse sensations; and (3) its probability of being perceived as a
metrical accent or downbeat (cf. Longuet-Higgins & Lee, 1982; Rosen-
thai, 1992; Thomassen, 1982; Vos, 1977).
Massaro (1970) demonstrated that the perceptual salience of a tone can
be reduced by presenting a second tone within the time span of echoic
storage. He concluded that the second tone interrupted the processing of
the first. This suggests that events followed by longer IOIs are more salient
than events followed by shorter IOIs. Povel and Okkerman (1981) found
this idea to be consistent with their findings on subjective accentuation.

assumption: Durational accent increases with IOI for small values of IOI
and saturates as IOI approaches and exceeds the duration of the echoic
store (auditory sensory memory).
In Experiment 1, it was observed that the downbeat of a cyclic pattern
(the event initiating the longest IOI) became less ambiguous as the rhythm
was played faster: At faster tempi, phenomenal accent was usually greater
for events preceding longer IOIs, whereas at the slower tempi, no consis-
tent relationship was evident between phenomenal accent and IOI. Mas-
saro's concept is consistent with these results, given that the IOIs in Experi-
ment 1 were predominately greater than the duration of the echoic store at
very slow tempi and less at very fast templ.
428 Richard Parncutt

Echoic storage has been referred to by many different names, among

them echoic memory (Neisser, 1967), precategorical acoustic storage
(Crowder & Morton, 1969), memory for nonattended auditory material
(Glucksberg & Cowen, 1970), preperceptual auditory store (Massaro,
1970), brief auditory storage (Treisman & Rostron, 1972), nonverbal
memory trace (Deutsch, 1975), and primary or immediate memory
(Michon, 1975). For the present purposes, these diverse terms are assumed
to refer to essentially the same kind of "memory." Note that echoic stor-
age is not really memory at all in the conventional sense of the word -

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

rather, it may be regarded as a lingering of sensory experience in the
absence of conscious cognitive processing. Michon (1975) described it as a
precognitive buffer store for incoming information.
Most estimates of the duration of the echoic store vary between 0.5 and
2.0 s (Crowder, 1969, 1970; Guttman & Julesz, 1963; Kubovy & How-
ard, 1976; Rostron, 1974; Treisman, 1964; Treisman & Howarth, 1959;
Treisman & Rostron, 1972). Other studies have arrived at both lower
estimates of the duration of echoic storage, in the vicinity of 250 ms
(Massaro, 1970, 1974; Plomp, 1964), and higher estimates such as 3 s
(Fraisse, 1982) and even 5 s (Glucksberg & Cowen, 1970).
The variability in published estimates of the duration of echoic storage
may be due to differences in experimental method. The duration of mem-
ory storage is somewhat flexible, depending on the kind of sound being
stored, and the context in which it is heard (Michon, 1978). Another
source of uncertainty is the arbitrariness of the point in time at which
memory is deemed to be exhausted. Suppose, for the purpose of argument,
that the salience of sensory material in echoic memory decays exponen-
tially with a half-life of 1 s (cf. Kubovy & Howard, 1976). The perceptual
salience of an event in the sensory store would then fall to 1/2 its original
value after a period of 1 s, 1/4 after 2 s, 1/8 after 3 s, and 1/32 after 5 s.
Which of these times represents the duration of echoic memory is a matter
of definition.
An alternative explanation for the relationship between accent and IOI
involves familiarity with the auditory environment. Consider an environ-
ment in which sound events (produced, for example, by collisions between
objects) are distributed at random time intervals of the order of 1 s. The
succession of sound events perceived by a human observer in such an
environment will be nonrandom, for the following reasons. First, rela-
tively loud sounds will tend to mask relatively quiet sounds that are either
simultaneous or close to them in time. Sounds that are completely masked
may be regarded as perceptually irrelevant. So perceived time intervals in
the vicinity of louder sounds will tend to be longer than perceived time
intervals in the vicinity of quieter sounds. Second, loud sounds often begin
with a fast, percussive onset, followed by a sustain/decay segment during
 Pulse Salience and Metrical Accent 429

which other, quieter sounds may be masked. The sustain segment is typi-
cally prolonged by the acoustic resonance of the sound source and by
reflection of the radiated sound from objects and surfaces in the vicinity.
So perceived time intervals following loud sounds will tend to be longer
than perceived time intervals preceding loud sounds. Moreover, forward
masking is stronger than backward (Moore & Glasberg, 1983; Zwicker &
Fasti, 1972), further enhancing the positive correlation between the IOI
following an event and the loudness of the event.
The saturation of durational accent for IOIs exceeding roughly 1 s may

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

be related to the typical or average duration of perceived IOIs that follow
loud sounds in the everyday auditory environment, including the sounds
of speech and music. Given that loud sounds tend to mask quiet sounds,
perceived IOIs may be influenced by physical parameters such as the decay
times of typical physical resonators and typical time intervals between
direct and reflected sound. The sum of these two contributions would
typically lie in the range of, say, 0.1-2.0 s, in both musical and nonmusical
auditory settings.
Yet another possible explanation for the saturation of durational accent
with increasing IOI is suggested by data on the perceived loudness of single
tones of various durations. Extrapolating from experimental results of
Plomp and Bouman (1959), Roederer (1979, pp. 88-89) pointed out that
the loudness of relatively short tones (lasting less than about half a second)
depends on their onset-to-offset duration, increasing steadily as duration
increases. The loudness of such tones may be understood to be determined
by the total physical power they deliver to the ear. For durations exceeding
about 1 s, loudness approaches a constant value that depends on intensity
(physical power per unit time) and is independent of duration. The loud-
ness of a single tone thus varies with its duration in much the same way as
the durational accent of a short rhythmic event varies with the IOI follow-
ing that event. Moreover, the loudness of higher-frequencytones saturates
more quickly than the loudness of lower-frequency tones, suggesting that
the saturation time for durational accent may depend similarly on the
pitch of the sound event that initiates the IOI.

assumption: The durational accent of small IOIs depends on an effective

IOI that is roughly 50-100 ms shorter than physical IOI.
In their system of rules for music performance, Friberg, Frydén, Bodin,
and Sundberg (1991) lengthened extremely short notes. The extra time
was borrowed from neighboring (longer) notes, hence the rule name social
duration. In addition, a minimum duration of 50 ms was specified.
The procedures adopted by Friberget al. may be interpreted as compen-
sating for the minimum discriminable IOI. Consider a typical isochronous
sequence. As the period of the sequence approaches 50-100 ms (10-20
430 Richard Parncutt

Hz), the events of the sequence eventually cease to be discriminable. At the

threshold of discriminability, the IOI perceived to follow each event will
effectively be zero. If the speed of the sequence is then gradually reduced so
that individual events again become audible, the durational accent of each
event will increase monotonically as the difference between the actual IOI
(or period) and the minimum discriminable IOI increases. In other words,
the relationship between durational accent and IOI will not pass through
the origin.
The minimum discriminable IOI appears to be closely related to the

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

perceptual onset delay of a sound. Morton, Marcus, and Frankish (1976)
defined the P-center of a speech vowel as the perceptual moment of occur-
rence of a word or phoneme. The P-center generally lags behind the physi-
cal onset of a sound by a time interval that may be related either to
acoustic factors, such as physical duration and amplitude envelope (How-
ell, 1988; Marcus, 1981), or to articulatory timing (Fowler, 1979). In the
case of musical tones, perceptual onset delay has been related to the shape
of the rise portion of the tone. Schutte (1978) proposed a temporal integra-
tion model, in which perceptual onset occurs when the area under the
subjective amplitude envelope passes a specific fraction of its maximum
value. Vos and Rasch (1981) suggested that perceptual onset of a tone
occurs when the amplitude envelope passes a threshold situated 6-15 dB
below maximum amplitude, for maximum amplitudes in the range of 20-
70 dB (respectively) above masked threshold.
For rhythmic sequences made up of physically identical sound events
(such as the stimuli in the above experiments), the perceptual onset of each
event is presumably delayed by a constant time interval. In such sequences,
perceived IOIs probably do not depend on perceptual onset delays, but
correspond exactly to physical IOIs. In very fast rhythms, however, the
minimum discriminable IOI may be influenced (or even determined) by the
perceptual onset delay of each event. This hypothesis is feasible given that
(1) perceptual onset delays and the minimum discriminable IOI typically
have about the same order of magnitude (some 50-100 ms), and (2) both
phenomena may be explained by a temporal integration process. In real
musical rhythms, of course, sound events are not physically identical, and
perceptual onset delays vary from one event to the next. Such variations
are not accounted for in the present version of the model.
In the present model, the above three assumptions are accounted for as
follows:
-
Ad(T) = [1 cxp{-ioi(T)/r}]\ (3)
where Ad(T) denotes the durational accent of the event occurring at time T,
exp is the natural exponential function, ioi (T) is the IOI following the
event expressed in milliseconds, r is the saturation duration (assumed
 Pulse Salience and Metrical Accent 43 1

proportional to the duration of the echoic store), and i is the so-called

accent index, accounting for the minimum discriminable IOI. The vari-
ables t and i are the first two free parameters in the model.
Note that Equation 3 could be expressed slightly differently by substitut-
ing 2* for exp (or e*). If i were equal to 1, then r would correspond to the
half-life of the echoic store. The natural exponential is preferredhere, as it
corresponds to standard mathematical usage.
Equation 3 is graphed in Figure 6 using parameter values found to pro-
duce a good fit between calculations according to the model and experimen-
tal results: r = 500 ms and i = 2 (see Table 1 below). The graph suggests that

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

durational accent saturates at an IOI of about 1 s, in agreement with most
experimental findings on the duration of the echoic store. The further one
moves to the left on the graph (i.e., the faster the tempo of a rhythm), the
greater the difference in phenomenal accent between long and short IOIs
(i.e., the greater the difference between the salience of corresponding pulse
sensations), in agreement with the results of Experiment 1.
The concave-upward shape of the left portion of the curve in Figure 6
accounts for the effect of minimum discriminable IOI. The effect could
have been accounted for more directly in Equation 3 by setting i = 1 and
-
replacing toi (T) by the expression max {ioi(T) ioio, 0}, where ioio is the
minimum discriminable IOI in milliseconds. This modification would have
the advantage that r does not interact significantly with ioio (whereas t
interacts with i in Eq. 3 as formulated above). However, the modification
would necessitate an extra routine to deal with events separated by an
interval of less than ioio. If two such physical events combined to form a
single perceptual event, when would its perceptual onset occur? In the
absence of experimental data on this issue, the parameter ioio was ex-
cluded from Equation 3.
An alternative possible explanation for the shape of the curve in Figure
6 involves metrical ambiguity. Increasing the value of the parameter i has

Fig. 6. Dependence of phenomenal accent on interonset interval (IOI) according to Equa-

tion 3 with t = 500 ms and i = 2.
432 Richard Parncutt

the effect of sharpening the perceptual difference between long and short
events, thereby sharpening the difference between the salience of different
pulse responses and making metrical interpretations less ambiguous. Ac-
cording to this explanation, i should increase with musical experience.
Does the IOI preceding an event influence its durational accent? Povel
and Essens (1985) proposed that an event in a nonisochronous sequence
of physically identical sounds is heard as accented, or perceptually
marked, if it satisfies one of the following three criteria: (1) it is relatively
isolated, (2) it is the second of a cluster of two events, or (3) it is first or

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

last in a cluster of three events. They then assigned subjective accents of
two units to events satisfying the above criteria, and one unit to others.
Consistent with Equation 3, point (2) implies that an event is accented if
the IOI that follows it is greater than the IOI that precedes it. Point (3)
additionally implies that, under certain conditions, the IOI preceding an
event may enhance its accent, which is inconsistent with Equation 3. Point
(3) was confirmed by Drake, Dowling, and Palmer (1991), who found
that the first and last events in a rhythmic group were accented in perfor-
mances of simple melodies by children and adults. Equation 3 is neverthe-
less consistent with Lee (1991), who compared various models of rhythm
perception and found that models in which accent was primarily deter-
mined by the IOI following an event performed considerably better than
others.
While formulating the model, attempts were made to incorporate the
effect of the IOI preceding an event on the event's durational accent, but
these amendments failed to produce any overall improvement in the corre-
lation between calculations and results of the experiments. This result may
simply have been due to the particular set of rhythmic stimuli used in the
experiments. The effect of the previous IOI may be incorporated into a
future version of the model by testing a wider range of rhythmic patterns,
such as those tested by Drake (1993), Povel and Essens (1985), or Sum-
mers, Hawkins, and Mayers (1986).
According to Povel and Okkerman (1981), events preceding longer IOIs
are more accented only for IOIs greater than about 250 ms whose relative
difference is greater than 12-15%. They found that if two successive
events are followed by small IOIs of almost the same magnitude, then the
sound followed by the shorter IOIs is perceived as accented. This effect
was not incorporated into the present model, for two reasons. First, this
kind of accent is quite weak (Povel, 1984). Second, IOIs in musical
rhythms that differ by less than 10-15% are generally perceived as cate-
gorically identical (Clarke, 1987; Fraisse, 1956). The present model is
limited to sequences in which IOI differences are perceived as categorically
different.
 Pulse Salience and Metrical Accent 433

PULSE PERCEPTION AS PATTERN RECOGNITION

assumption: The perception of pulse involves the spontaneous recogni-

tion of sequences that are perceived (categorically) to be isochronous.
A pattern-recognition approach to pulse perception was described by
Parncutt (1987) and Rosenthal (1992). In everyday visual scenes, a famil-
iar pattern may be recognized even if parts of the pattern are missing, for
example if a close object obscures part of a distant object. In the case of
pulse perception, recognition is possible even if elements of the pulse are
missing- for example, if there is a rest at the start of a measure of music,

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

or if part of a pulse train performed by one instrument is masked by
another instrument. Musical pulse sensations are generally illusory in that
they do not necessarily indicate the presence of single periodic sound
sources. In this regard, a pulse sensation is analogous to a musical chord
that is perceived to blend or fuse into a single sound, even though it was
created by three or more different sources.
The perception of pulse trains in musical rhythms may be conceptual-
ized as a process by which sound onsets are matched against the elements
of an isochronous template. A given template may be fully specified by
two pieces of information: its period P and phase Q. In the present model,
both P and Q are measured in basic time units b.
Phase is defined as the temporal position of an event in a pulse relative
to some arbitrary reference time, such as the nominal start of a given
rhythmic cycle. At the reference point, phase Q is set equal to 0 (instead of
1, for the first beat of the bar), as this simplifies the mathematics. The
waltz rhythm of Figure 1, for example, evoked pulse sensations of period
3, both "on the beat" (beat = 1, phase = 0) and "off the beat" (beat = 3,
phase = 2).

definition: The salience of a pulse sensation is a measure of the probabil-

ity that a listener will tap out that pulse when asked to tap the tactus (main
or underlying beat) of a rhythmic sequence.
This operational definition of pulse salience corresponds to the proce-
dure of Experiment 1. The definition is reminiscent of the psychoacoustic
definition of the pitch of a complex sound as the frequency of a pure tone
judged to have the same pitch (American Standards Association, 1960). It
is also reminiscent of the definition of pitch salience as the probability that
a listener will select a pure tone of the corresponding frequency in a pitch-
matching experiment (cf. Parncutt, 1989; Terhardt, Stoll, & Seewann,
1982).
The definition of pulse salience may be made more precise by requiring
that listeners tap with their dominant hand and that they use the index
434 Richard Parncutt

finger, middle finger, or both together. This additional specification is not

crucial, as rhythmic imitation performance is largely independent of motor
constraints such as specific combinations of hands and fingers (Keele,
Pokony, Corcos, & Ivry, 1985; Michon, 1967; Mishima, 1965; Summers,
Bell, & Burns, 1989).

assumption: Each pair of events in a rhythmic sequence initially contrib-

utes to the salience of a single pulse sensation, in proportion to the product
of their phenomenal accents; and contributions to the salience of a pulse

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

sensation from different pairs of events add linearly.
This principle, proposed by Parncutt (1987), produces sensible results
when incorporated into the present model. The word "initially" alludes to
the assumption (below) that pulse sensations can enhance the salience of
other, consonant pulse sensations.
According to the above assumption, as few as two events can produce a
sensation of pulse. This conflicts with Krebs' claim (1987, footnote 16)
that at least three regularly recurring events are needed to evoke a level of
motion. Krebs pointed out that two equal IOIs are needed to imply
isochrony, and that three events are needed to define two IOIs. According
to this line of reasoning, a sequence of alternating 1/2 and 1/4 notes (the
waltz sequence in Figure 2b) would never imply a pulse with a period of
1/4 note. Given that many listeners in Experiment 1 tapped at 1/4 note
intervals in response to the waltz sequence, Krebs' hypothesis may be
rejected in favor of the principle quoted above.
It is possible that the 1/4 note responses to the waltz rhythm in Experi-
ment 1 were artifacts of the experimental method (Drake, personal com-
munication). Listeners may have experienced no pulse sensation at all in
response to waltz and chosen the 1/4-note pulse simply because it was the
easiest to tap out. However, all the stimuli in that experiment were played
at tempi found in real music, suggesting that all evoked at least one pulse
sensation; and listeners were free to tap out any pulse sensation they
wished. In any case, if the above definition of pulse salience is valid, then
pulse salience is defined by the results of Experiment 1.
The above assumptions are incorporated into the following formulation
of the goodness of fit, or pulse-match salience S'm,of a pulse of period P
and phase Q:

S;= 2 Ao (modc {nP + Q})Ao (modc {[» + 1]P + Q}), (4)

n=O,N-l

where N = 1cm {P, C}.

Here, 2 denotes summation, Ao is phenomenal accent (Eq. 2), modc
means "modulo base C" (e.g., mod3, in the case of waltz in Figure 1), lem
stands for lowest common multiple, and N is the minimum number of
 Pulse Salience and Metrical Accent 435

summation terms needed to estimate the salience of a pulse of period P

evoked by a rhythm of cycle length C.
Povel and Essens (1985) estimated the salience of a pulse sensation
(which they called the strength of an induced clock) by a method similar to
the above pattern-matching procedure. A difference between the two for-
mulations is that theirs was based on the amount of counterevidence a
clock meets in an actual sequence- the number of clock ticks falling on
unaccented or missing events (called ghost events by Rosenthal, 1992).
Later in the same paper, they suggested that clocks are actually induced by

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

positive evidence, that is, by accented events, as in the present model.
Rosenthal (1992) determined pulse sensations (rhythmic levels) by a
method similar to the present one, but did not explicitly evaluate a salience
value for each pulse candidate.
It may be possible to improve Equation 4 in a future version of the
model by reference to data on the discrimination threshold for isochrony-
the sensitivity of listeners to small deviations from temporal regularity-
on the assumption that this threshold is proportional to the salience of the
corresponding pulse sensation (see the section below entitled Temporal
precision and discrimination). Schulze (1989) investigated the discrimina-
tion threshold for isochrony as a function of the number of isochronous
events in a sequence, for pulse periods in the range 50-300 ms. Sequences
deviated from physical isochrony only at the last event. The threshold
decreased as the number of events increased, approaching a minimum
value of roughly 10 ms for sequences comprising more than about five
beats (cf. 20 ms in musical contexts: Clark, 1989). Drake and Botte (in
press) presented sequences in which the only deviation from isochrony was
a slight change of tempo near the position of a missing event at the middle
of the sequence. Their results imply that the relative just noticeable differ-
ence (JND) approaches an asymptotic minimum of 2-3% for quasi-
isochronous sequences comprising four to eight sound events (referred to
in the paper as two to four intervals). The relative JND was roughly
constant over a wide range of periods (200-1000 ms) and sank as low as
1.5% for sequences of 8-12 events (four to six intervals) in the range 300-
800 ms.
The next stage of the model involves the normalization of pulse-match
salience so that it equals one in the simple case where the stimulus is an
isochronous sequence and the pulse sensation corresponds exactly to the
stimulus. In the experiments described above, this occurred when pattern
(a) pulse was presented at a slow tempo, and listeners tapped at the same
rate as the stimulus (i.e., without grouping). The purpose of the normaliza-
tion is to remove any tempo dependence from calculated pulse saliences.
This makes it easier to introduce explicit tempo dependence (Eq. 6). The
normalization is effected as follows:
436 Richard Parncutt

C'
m N V'
(1 - exp{-/?/r}P
where N is defined as in Equation 4, and p = Pb is the pulse period in
milliseconds. (The expression in parentheses is derived from Eq. 3.)

MODERATE TEMPO AND THE EXISTENCE REGION OF PULSE

SENSATION

The most salient pulse sensation (tactus) evoked by a rhythm depends

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

on its tempo (Handel, 1986; Handel & Oshinsky, 1981). The present
model formulates this dependency quantitatively by means of a function
called the existence region of pulse sensation.

definition: The existence region of pulse sensation is a range of periods

within which isochronous sequences are perceived to be musically rhyth-
mic (or to imply movement).
assumption: The median, or (logarithmic) center, of the existence region
of pulse sensation corresponds to a moderate musical tempo. In general,
the closer the tempo of a pulse to moderate tempo, the greater the salience
of the corresponding pulse sensation.
Jones and Boltz (1989) referredto a dominance region that "reflectsnatu-
ral attending limits that are independent of ratio-complexity constraints . . .
Specifically,the dominance region involves a set of relatedtime levels around
the referent beat period . . . Dominant region limits are relative not absolute
time limits in that they are expressed in terms of hierarchicallevels" (p. 473).
The present concept of existence region of pulse sensation is similar to the
concept of dominance region of Jones and Boltz, with the exception that the
existence region is assumed to depend directly on absolute period.
The existence region of pulse sensation has an interesting analog in the
area of pitch perception. Plomp (1967) and Ritsma (1967) found that the
pitch of complex tones in speech and music is usually determined by
harmonics other than the fundamental. Their results implied that the de-
gree to which a given harmonic influences the overall pitch of a complex
tone depends on both its absolute frequency and its harmonic number.
Terhardt et al. (1982) treated effects of absolute frequency and effects of
harmonic number as independent, associating effects of harmonic number
with a pattern-matching (subharmonic coincidence) routine, and effects of
absolute frequency with a bell-shaped spectral frequency weight function.
The existence region of pulse sensation is similar to the spectral frequency
weight function in that it is assumed to depend only on absolute period.
The shape and extent of the existence region of pulse sensation region
may be determined in three different ways.
 Pulse Salience and Metrical Accent 437

1. Spontaneous, personal, motor, or mental tempo. Subjects tap

out isochronous sequences (Braun, 1927; Fraisse, 1963;
Fraisse, Pichot, & Clairouin, 1949; Guttmann, 1931; Harrell,
1937; Harrison, 1941; Meumann, 1894; Miles, 1937; Miy-
ake, 1902; Patterson, 1916; Reymert, 1923; Rimoldi, 1951;
Seashore, 1899; Squire, 1901; Stern, 1900; Wu, 1935). Ide-
ally, no prior suggestion is made about tempo; alternatively,
tempo may be specified as moderate.
2. Preferred tempo. Pulse trains (e.g., metronome ticks) are

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

played to listeners, who then indicate which rates they preferor
find most natural (Frischeisen-Kôhler, 1933; Wallin, 1901,
1911,1912).
3. Tactus. Listeners tap out the tactus (main or underlying beat)
of a musical excerpt or rhythm, as in Experiment 1 (see also
references on subjective rhythmization under Enhancement
of pulse salience).

In each case, responses are limited to much the same range of periods.
Perusal of the experimental literature permits the following generaliza-
tions (cf. Fraisse, 1982).

1. The most salient pulse sensations have a moderate tempo of

about 100 isochronous events per minute, or a period of
about 600 ms.
2. The region of greatest pulse salience (dominance region) lies
between about 2/3 and 3/2 times moderate tempo: 67-150
events per minute, or 400-900 ms.
3. Pulse sensations cease to exist beyond about 1/3 or three
times moderate tempo: 33-300 events per minute, or 200-
1800 ms (cf. Bolton, 1894; Fraisse, 1956, 1982; Handel &
Lawson, 1983; MacDougall, 1903; Michon, 1978).

Regarding the lower limit on the tempo of a pulse sensation, Lerdahl

and Jackendoff (1981) pointed out that pulse sensations whose periods
exceed two written measures are generally imperceptible. Such levels may
be regarded as belonging to the domain of musical form rather than
rhythm. Regarding the upper limit, the literature suggests that isochronous
sequences faster than about 300 events per minute may sound rhythmic,
but the pulse sensations produced by such sequences probably have peri-
ods corresponding only to multiplies of the period of the original signal
(subjective rhythmization) rather than the period itself.
The dependence of pulse salience on tempo is accounted for in the
model by a pulse-period salience function Sp, a normal or Gaussian (bell-
438 Richard Parncutt

shaped) function of the logarithm of the pulse period p similar to that

displayed in Figure 3 above:
Sp= expHOg10(£)]2}. (6)
Here, /x denotes moderate pulse period and is typically about 600 ms; and
o"is the standard deviation of the logarithm of pulse period, a measure of
the width of the existence region, and typically has a value of about 0.2.
The variables [x and a are the third and fourth free parameters of the
model. The function may be regarded as a kind of band-pass filter, admit-

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

ting only pulse sensations that lie within a given range of periods.
The bell-shaped function defined in Equation 6 is quite broad, with a
relatively poorly defined peak. The model is thus relatively unaffected by
small variations in jjland a. Most published experimental results on spon-
taneous motor tempo, the tempo of pulses perceived in response to
rhythms, and preferred tempo are consistent with this function.
Fraisse (1956, 1982) asked people to tap at equal time intervals and to
group the taps into threes or fours. He reported that intertap periods were
considerably shorter than the usual 600 ms, averaging 420 ms (for groups
of three) and 370 ms (for fours). Similar results had been obtained in
earlier studies (MacDougall, 1903; Miles, 1937; Miyake, 1902; Wallin,
1901). These findings may be explained with reference to the existence
region of pulse sensation. When subjects are asked to group taps into
threes or fours, they presumably choose tempi that will maximize the
aggregate salience (rhythmicity?) of all pulse sensations evoked by the
sequence. The most salient pulses in Fraisse's experiment presumably cor-
responded to the tapping period and the group length. Due to the concave-
downward (bell) shape of the existence region (Eq. 6), the sum of the
saliences of these two pulses is maximal when the two periods lie on either
side of peak of the curve- that is, when one pulse has a period of consider-
ably less than 600 ms, the other considerably more than 600 ms.
The existence region of pulse sensation is assumed here to be a continu-
ous function of pulse period. A recent study of synchronization (Collyer,
Broadbent, & Church, 1992) suggests that this assumption may not be
entirely valid. In that study, listeners tapped in time with an isochronous
sequence and continued tapping after the stimulus was turned off. System-
atic deviations from the original tapping periods during the continuation
phase suggested the existence of temporal categories in motor control.
This effect seems unlikely to affect musical applications of the present
model.
assumption: The above two contributions to pulse salience- goodness of
fit between the rhythm and the corresponding isochronous template, and
the tempo of the pulse - are essentially independent of one another.
 Pulse Salience and Metrical Accent 439

The overall salience S of a pulse sensation is estimated in the model

by multiplying its pulse-match salience Sm by its pulse-period salience
v
S = [SmSpl (7)
The symbol / is the so-called the pulse-salience index, the model's fifth
and final free parameter. Like i in Equation 3, / generally exceeds one. It
exaggerates differences between pulse saliences, thereby reducing the ambi-
guity of the tactus of a rhythm, and may therefore be expected to increase

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

with listeners' confidence about the location of the underlying beat of a
rhythmic sequence.
The salience of pulse sensations may also be affected by motivic repeti-
tion (Lee, 1991; Lerdahl & Jackendoff, 1983; Longuet-Higgins & Lee,
1982, 1984; Steedman, 1977). The repetition of temporal patterns may be
accounted for systematically in a model of rhythm perception by the
mathematical technique of autocorrelation (Brown, 1993). However, no
effect of motivic repetition was evident in the present analysis of the results
of Experiment 1: The salience of pulse sensations whose period corre-
sponded to the cycle length did not consistently exceed predictions (see
Figure 2). Perhaps the effect of motivic repetition only affects pulse percep-
tion in more interesting (noncyclic) rhythmic patterns or in melodic mo-
tives that are defined by both rhythm and contour.

METRICAL ACCENT

According to Lerdahl and Jackendoff (1983, p. 17), "Phenomenal ac-

cent functions as a perceptual input to metrical accent." The present
model clarifies this relationship by breaking it up into two stages. First,
phenomenal accents (here restricted to durational accents) determine the
salience of pulse sensations. In the second stage, pulse saliences determine
the strength of metrical accents (Parncutt, 1987).
Longuet-Higgins and Lee (1984) have noted that "The weight of a given
note or rest is the level of the highest metrical unit that it initiates" (p.
430). According to Rosenthal (1992), the salience of an event is "an
appropriate tally of properties of the note that make it more likely to be
considered a downbeat" (p. 68). In the present model, metrical accent is
estimated according to the following principle.

assumption: The metrical accent of an event or temporal category {includ-

ing a missing event such as a rest at the start of a bar) may be estimated by
combining the saliences of all pulse sensations including that event or
temporal category.
440 Richard Parncutt

Specifically, metrical accent Amat time T is estimated by linear addition

of the saliences of all the pulse sensations converging on time T:

K(T)=^Sp(P,Q) (8)

where modc{Q + nP} = T for some integer n.

In practical applications of Equation 8, computation time may be saved
by limiting pulse periods to a maximum of about 2000 ms. Note also that
results as a function of phase Q repeat cyclically with a period of hcf{P, C},

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

where hcf stands for highest common factor. Computation time may be
saved by restricting calculations to phases Q in the range 0 to hcf{P, C} - 1.
The described concept of pulse sensation includes hypermeters, that is,
pulse sensations whose periods exceed that of the notated meter (Roth-
stein, 1989). So the above formulation of metrical accent also includes
Rothstein's concept of hyperaccent. In the present approach, hypermeters
must be fast enough to contribute directly to the music's rhythmic feel
(that is, more than 30-40 beats per minute) if they are to be regarded as
pulse sensations. Slower hypermeters may subdivide the music into formal
chunks that happen to be of approximately equal length, but do not imply
rhythmic movement.

MODELING EXPERIMENTAL RESULTS

The above model was tested by comparing calculations with the results
of the experiments reported above. Values for the free parameters of the
model were estimated by an automatic procedure implemented in the
language Le_Lisp. The procedure adjusted parameter values by small steps
until the Pearson correlation coefficient r between measurements and cal-
culations was a maximum. The results of this procedure are shown in
Table 1. These results were independent of initial parameter estimates for
a range of feasible sets of initial values.
In Experiment 1 (pulse salience), measurements were compared with
calculations according to Equation 7. Computation time was reduced by
considering only those pulses (periods, phases) that had been selected in the
experiment (161 data points). Calculated values were first normalized so
that, for each trial of the experiment, their sum (like the sum of the experi-
mental data) was 22, the number of listeners in the experiment. Then all five
free parameters (r, /, )Lt,cr,/')were gradually and independently varied until
the correlation between calculations and experimental data was optimal (r
= .88). This value is high in view of the relatively large number of degrees of
freedom in the comparison: 161 (data points) - 36 (normalization) - 5
(free parameters) - 2 (correlation) = 118. The calculated pulse saliences
from which this value was determined appear in italics in Figure 2.
 Pulse Salience and Metrical Accent 441

TABLE 1
Optimal Values of the Free Parameters
Parameter t i n a j r

1. Pulse salience
a. Individual pulses 550 1.6 760 0.23 2.0 0.88
b. Distribution of periods - - 710 0.22 - -
2. Accent
a. Phenomenal 200 2.0 - - - 0.87
b. Metrical 180 1.7 660 0.14 1.9 0.95
3. "Typical" 500 2 700 0.2 2 -

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

note. - Definition of parameters: r is the saturation duration, or the duration of the
echoic store, in milliseconds (Eq. 3); /is the accent index (Eq. 3); y. is the moderate pulse
period in milliseconds (Eq. 6); cr is the width of the existence region of pulse sensation,
expressed as a standard deviation of log periods (Eq. 6); / is the pulse index (Eq. 7); and r
denotes the coefficient of correlation obtained in each case between calculations and experi-
mental results, by using the listed parameter values.

The optimal values of the free parametersfor Experiment1 (Table1,

line la) make sense in terms of their interpretationsin the model. In
particular,the value of the optimalpulse period found by this procedure
(jit= 760 ms), and the relativelylargevalue for the standarddeviationof
its logarithm(a = 0.23), are consistentwith most literatureon the center
and width of the existence region of pulse sensation,and the saturation
durationr (550 ms) is consistentwith most literatureon the durationof
the echoic store.
The 26 results of Experiment2 (metricalaccent) were compareddi-
rectly with calculatedmetricalaccentsaccordingto Equation8. The re-
sults are shown in line 2b of the table. Consistentwith the lower number
of degrees of freedom in this comparison [26 (data points) - 5 (free
-
parameters) 2 (correlation)= 19], the correlationcoefficientwas higher
than before (r = 0.95). The optimal values of the parametersi, fi, and ;
were about the same as for Experiment1. The saturationdurationr and
existence-regionwidth a were considerablysmallerthan in Experiment2,
-
presumablybecauseof the restrictedrangeof tempiin that experiment
all rhythmswere playedat the samerateof 150 notesperminute.The IOIs
in Experiment1 morethoroughlycoveredthe durationalaccentsaturation
curve(Eq.3) and the pulsesensationexistenceregion(Eq.6); so the values
of r and a estimatedfromthe resultsof Experiment1 are morelikelyto be
musicallyrepresentativeor typical.
The results of Experiment2 were also compareddirectlywith dura-
tional accentsaccordingto Equation3, in orderto obtain additionalesti-
mates of the first two free parametersr and i (line 2a in Table 1). The
resultantcorrelationcoefficientwas considerablylower for phenomenal
than for metricalaccent (.87 versus .95), suggestingthat the measured
442 Richard Parncutt

event saliences were indeed metrical rather than phenomenal in nature-

in accordance with the instructions given to listeners in the experiment,
which made direct reference to downbeats and syncopation.
The "typical" values of the free parameters in the bottom line of Table 1
are intended for music-theoretical applications of the model.

Extensions and Applications

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

The model as set out and tested above is sufficiently coherent, parsimo-
nious, and accurate for a range of music-theoretical applications. How-
ever, it could be changed in many ways to satisfy future requirementsor as
new theoretical possibilities emerge. The model may also be extended to
account for a range of rhythmic phenomena not dealt with in the de-
scribed experiments. Some possible extensions are set out below.

RHYTHMIC CONSONANCEAND PERCEIVEDMETER

Yeston (1976) described two strata (pulse sensations) as rhythmically

dissonant if their note values were not whole multiples or factors of each
other. Krebs (1987) referred to such a relationship as a type A dissonance,
and went on to define type B dissonances as nonaligned (or displaced)
combinations of pulse trains whose periods were whole multiples or factors
of each other. These definitions were based on Hlawicka's (1958) distinc-
tion between Verwechslungen and Verschiebungen, respectively. Krebs
cited several examples of type A and B dissonances from Western art music.
Common type A dissonances are polyrhythms such as 2 against 3, 3
against 4, and so on. Studies on the perception of polyrhythms (Beau-
villain &cFraisse, 1984; Handel & Oshinsky, 1981) have shown that one
of the two dissonant pulses can act as a perceptual frame of reference for
the other, but that this happens only at relatively fast tempi- specifically,
for reference pulse periods smaller than about 500 ms. This finding may be
explained in terms of the existence region of pulse sensation. Consider, for
example, a 2 x 3 cross rhythm of 1/4 and 1/6 notes. The superposition of
these two pulses may produce an additional pulse sensation of 1/12 notes,
here called an interaction pulse. For one of the two main pulse trains to be
perceived as a frame of reference, it should evoke a considerably stronger
or more salient pulse sensation than the interaction pulse. Because of the
bell shape of the existence region of pulse sensation, this is possible only if
the period of the reference pulse is smaller than the moderate pulse period
(600-700 ms), so that both reference and interaction pulses lie on the
same (left) side of the peak. If this is the case, the interaction pulse will be
much weaker than the reference pulse.
 Pulse Salience and Metrical Accent 443

assumption: A prerequisite for the perception of meter is the concurrent

perception of consonant pulses.
According to Yeston (1976), meter may be regarded as an outgrowth of
the interaction between two or more consonant pulse sensations (or rhyth-
mic strata). For example, a 3/4 meter may be specified by a pulse of 1/4 notes
(the beats) and a pulse of 3/4 notes (the measures or downbeats), where
every third 1/4-note coincides with a 3/4-note. This combination satisfies
Lerdahl and Jackendoff's (1983) metrical well-formedness rule no. 2 (p.
69): "Every beat at a given level must also be a beat at all smaller levels."

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

A consonant set of pulses may be conceived as a hierarchical structure.
Lerdahl and Jackendoff rreferredto specific elements of hierarchical struc-
tures, including consonant rhythmic levels, as subordinate or superor-
dinate to each other. Rosenthal (1992) used the equivalent terms parents
and children (for pulses two or three times slower or faster than a given
pulse), and ancestors and descendants (for more distant consonant pulses).
For example, the 1/4-note pulse in a 3/4 meter is a child of the 3/4-note
pulse; an 1/8-note pulse would be a grandchild. Following Rosenthal's
terminology, a perceived meter may be regarded as a family of consonant
pulse sensations.

hypothesis: The salience of a perceived meter, or the probability of a

given metrical interpretation, is proportional to the sum (or some other
aggregate) of the salience of the pulse sensations that make it up. The most
likely meter to be perceived is the one with the highest predicted salience.
This aspect of the model has not yet been systematically tested. An
example of the procedure using a simple summation is shown in Parncutt
(1987); see also Rosenthal (1992).

ENHANCEMENT OF PULSE SALIENCE

hypothesis: Thé salience of a pulse sensation may be enhanced by the

presence of a parent or child pulse sensation, that is, a consonant pulse
sensation two or three times slower or faster. In other words, there is
mutual salience enhancement among consonant pulse sensations.
This hypothesis is capable of explaining the following two rhythmic
phenomena. First, in subjective rhythmization, subjective grouping by
fours is more common than grouping by threes. Subjective rhythmization,
or the perceptual grouping of isochronous sound events at moderate to fast
tempi, has a long history in experimental psychology (Bolton, 1894; Duke,
Geringer, & Madsen, 1991; Fraisse, 1956; MacDougall, 1903; Mach,
1886; Meumann, 1894; Miner, 1903; Miyake, 1902; Temperley, 1963;
Vos, 1973; Woodrow, 1951). Subjective rhythmization appears to involve
both serial and periodic grouping, where the two types of grouping are in
444 Richard Parncutt

phase- the start of each serial group corresponds to a periodic group

(downbeat). In the case of grouping by fours, subjective accents occur
twice per group, on the first and third elements. This implies that subjective
grouping by fours involves a family of three consonant pulse sensations, of
periods 1, 2, and 4 beats- as in 4/4 meter- whereas subjective grouping
by three involves only two consonant pulse sensations, of periods 1 and 3
beats- as in 3/4 meter. The enhancement of pulse salience due to the
presence of other, consonant pulse sensations would therefore be expected
to be greater in the case of subjective grouping by fours, making grouping

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

by fours more likely than grouping by threes (as in Experiment 1).
The predominance of binary/quaternary over ternary grouping is ex-
plicit in the rhythmic theory of Lerdahl and Jackendoff (1983) and evident
in experimental data obtained from both adults and children by Drake
(1993). In the present model, the predominance of binary groupings is
implicit: It is not stated directly, but arises from a combination of two
related assumptions. The first assumption is that pulse sensations are
confined to the existence region of pulse sensation. The number of conso-
nant pulse sensations that may exist simultaneously within this region is
greatest if the relationship between the pulses is binary (e.g., periods 200,
400, 800, 1600 ms). The second assumption is that consonant pulse sensa-
tions enhance one another.
An alternative explanation for the predominance of binary or quater-
nary groupings over ternary groupings is cultural conditioning. Binary
groupings (2/4, 4/4 meter) are more common in our musical culture, and
hence more familiar, than ternary groupings (Apel, 1970). The above
account sheds some light on the origin of this difference.
The second phenomenon that may be explained by pulse salience en-
hancement is the following. A sequence of alternating IOIs in the ratio 2:1
appears to evoke a pulse sensation of period 1, but alternating IOIs in the
ratio 3:1 (e.g., a dotted rhythm) do not. In other words, an isochronous
sequence may be perceptually subdivided by a single additional event into
three parts (2:1), but not four or more parts (3:1 or more). This claim is
supported by two pieces of evidence, stemming from research on the cate-
gorical perception of rhythmic patterns and performance timing.
Regarding categorical perception, Fraisse (1956) found that listeners
generally distinguished only two time categories in musical rhythms- long
and short, in the perceived ratio 2:1 - and categorized IOIs as the same
(i.e., both long, or both short) if the ratio of their IOIs was less than about
1.55:1. He also found that sequences of 2:l-related IOIs were judged more
rhythmic than other sequences. Such sequences are easier to encode and to
reproduce than other sequences (Drake & Gérard, 1989; Essens & Povel,
1985; Povel, 1981; Summers et al., 1989). Summers et al. (1986) found
that musicians could reproduce repetitive patterns of two IOIs specified
 Pulse Salience and Metrical Accent 445

only by their ratios, but that 2:1 was producedmoreaccuratelythan 3:1,

3:1 than 3:2 and 4:3, and 4:1 than 3:2.
Regardingperformancetiming, the performancerules of Sundberg
(1988) imply that alternatingIOIs (notated)in the ratio 3:1 sound more
musicalor naturalif the contrastbetweenthe two IOIs is heightened,or
madeto exceed 3:1; whereasalternatingIOIsin the ratio 2:1 sound more
musical or natural if the contrast between the two IOIs is softened, or
made smaller than 2:1. Similarconclusionsmay be drawn from earlier
measurementsof performancetiming (e.g., Gabrielsson,Bengtsson, &

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

Gabrielsson,1983; Seashore,1938).
The differencebetween2:1 and 3:1 ratios in alternatingIOI sequences
may be accountedfor as follows.

hypothesis: The salienceof a pulsesensationin whichno more than two

successiveeventsmatch real eventsis negligibleunlessthe pulse sensation
is enhancedby a salientparentor childpulsesensation.
Accordingto this hypothesis,alternatingIOIs in the ratio 2:1 imply a
pulse of period 1, supportedby the parentpulseof period3; but alternat-
ing IOIs in the ratio 3:1 do not imply a pulse of period 1 (as the present
modelwould predict),as thereis a gap of two generationsbetweenperiod
1 and period4. Forexample,a dottedrhythmof alternating3/8-notesand
1/8-notes does not, accordingto this hypothesis,imply a pulse of 1/8-
notes, unlessone or morerhythmicallyconsonant1/4-notesare presentin
the immediatecontext (bridgingthe gap between the two outer pulse
"generations").In the absenceof 1/4-notes,the fastestpulse sensationin
this case is probablythe 1/2-notepulse.
The modelusedto simulatethe resultsof the aboveexperimentsignored
this effect and may thus have predictedsome pulse sensationsthat are
actually imperceptible.However, the salience of such pulse sensations
accordingto Equation4 is generallyquite low, due to the relativelylow
phenomenalaccentassignedto shorterIOIsby Equation3.
The model of Povel (1981) allowedonly two differentsubdivisionsof a
pulse period: a numberof equal intervals,or two intervalsthat relate as
2:1. Subdivisionsin the ratio 3:1 were disallowed, unless equal (1:1)
subdivisionswere also prsent (e.g., 2:1:1, or 3:1:2:2). These restrictions
have essentiallythe same effect as the above hypothesis.The above hy-
pothesis has the advantage that it relates the effect of allowable beat
subdivisionsto the moregeneraleffectof pulse salienceenhancement.

CATEGORICAL PERCEPTION

Rhythmperceptionis categorical(Sloboda,1985): Soundsheard in a

metricalcontext are perceivedto correspondto specific (notable)beats
446 Richard Parncutt

(Povel, 1981). As discussed earlier, the distinction between even and un-
even subdivisions of a timespan is also perceived categorically (Clarke,
1987; Schulze, 1989).
The assignment of rhythmic events to temporal categories is a complex
process. For example, Clarke (1987) investigated the position of the cate-
gory boundary between alternative interpretations of rhythmic figures
(such as dotted, ternary, and binary subdivisions of a beat) and found that
the boundary tended to shift so as to favor perceptual judgments that
conform to the prevailing metrical context. Desain and Honing (1989)

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

modeled the categorization of musical time by a connectionist system.
The rhythms tested in the present study comprised IOIs in metronomi-
cally exact, small-integer ratios, and the model presented above was tested
only against the results of these experiments. Given the categorical nature
of rhythm perception, the model may be assumed also to apply to freer or
more musical renditions of simple rhythms, provided that temporal devia-
tions remain within category boundaries. It is nevertheless possible that
small deviations from mechanical timing may affect pulse salience. Effects
of this kind may be accounted for explicitly in an event-based model such
as Desain (1992). The present, pulse-based model does not account for
such effects.
In the above formulation of the model, it was assumed that every event
in a musical rhythm corresponded to a temporal category. Essentially the
same assumption was made by Lerdahl and Jackendoff (1983): "Every
attack point must be associated with a beat at the smallest level of metrical
structure." (Metrical Well-Formedness Rule no. 1, p. 69). However, two
phenomena - the tendency for listeners to perceive all IOI ratios as either
1:1 or 2:1 (Fraisse, 1956), and idiosyncratic differences between the tim-
ing of 2:1 and 3:1 ratios in musical performance (Sundberg, 1988) -
suggest that Lerdahl and Jackendoff's assumption may be invalid. Given
that an alternating sequence of IOIs in the ratio 2:1 can evoke a pulse of
period 1 unit but a sequence in the ratio 3:1 cannot (unless elements of IOI
= 2 units are also present), and that the event of unit IOI has metrical
accent in the 2:1 case but not in the 3:1 case, it is reasonable to hypothe-
size that the shorter event corresponds to a temporal category in the 2:1
case but not in the 3:1 case.

hypothesis: There is a one-to-one correspondence between metrical ac-

cents and temporal categories.
Consider two sequences of alternating IOIs, one in the nominal ratio
3:1 (notated as a dotted rhythm) and the other in the nominal ratio 7:1
(double-dotted rhythm). The previous hypothesis on pulse salience en-
hancement implies that, for both of these sequences, the shorter IOI has no
metrical accent. The present hypothesis further suggests that, in both
 Pulse Salience and Metrical Accent 447

cases, the shorter IOI is not perceivedas a temporalcategory.In other

words, there is no categoricalperceptualdifferencebetween cyclically
repeatingdotted and double-dottedrhythms.Double-dottedrhythmsare
of coursedistinguishablefromdottedrhythms(theysound "sharper"),but
the differenceis perceivedon a continuousscale.

EXPRESSIVE TIMING

Clarke(1987) distinguishedbetweenmeterand beat subdivision,which

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

areperceivedand performedcategorically,and expressivetransformations
(temporaldeviationsfrommechanicaltiming,or metronomicnote values),
which are continuous. Expressivetransformationsdepend in a complex
fashion on the nominal pitches and time values of a musical score. The
quantizationof eventsaliencein the presentmodelmay serveas a basis for
a new approachto the simulationof expressivetransformations.

hypothesis: Expressivetimingmay be modeledby a gradualslowing of

local tempo in the vicinityof metricaland structuralaccents.The amount
of slowing in the vicinity of an accent increasesas the strength of the
accentincreases.
Musical experiencesuggeststhe following relationshipbetween exact
musicaltiming and event salience:In musicalperformance,importantor
remarkableevents (e.g., eventsprecedingrelativelylong IOIs,on rhythmi-
cally strongbeats, at the startsof phrases,at harmonicdissonances,or at
phraseor structuralboundaries)may be slightlydelayedrelativeto metro-
nomic tempo by micropauses(cf. Bengtsson,1987; Bengtsson& Gabri-
elsson, 1983; Gabrielsson,1974; Palmer,1989; Repp, 1990; Seashore,
1938; Shaffer & Todd, 1987; Sundberg,Askenfelt, & Frydén, 1983;
Thompson,Sundberg,Friberg,& Frydén,1989). Usually,the moreimpor-
tant the phrase or structuralboundary,the longer the micropause.In
otherwords, the durationof the pausedependson the perceptualsalience
of the event starting the next phrase. This may be estimated from the
number of hierarchicallevels whose boundariescoincide at that point
(Todd,1985).
Researchon deviationsfrom metronomictempo has also shown that
relativelyimportanteventsin a musicalpassagetend to be lengthened,and
playedmore loudlyor morelegato (Clarke,1988; Shaffer,1981; Sloboda,
1983; Stetson, 1905; Todd, 1985; Vos, 1977). In musicology,lengthening
a note beyondits nominalIOI (i.e., beyondits notatedvalue)is said to give
the note agogic accent (Riemann, 1884). Agogic accent should not be
confusedwith durationalaccent(as definedin Eq. 3), which is assumedto
dependonly on physicalIOI, regardlessof nominalIOI (i.e., regardlessof
musicalnotation and context).
448 Richard Parncutt

Important events are thus both delayed and lengthened in musical per-
formance. This is equivalent to saying that local tempo (otherwise known
as instantaneous, or micro, or linear tempo) slows in the vicinity of impor-
tant events.
A continuous function of local tempo against time may be formulated
mathematically by analogy to the velocity and momentum of moving
objects (Kronman &cSundberg, 1987; Sundberg & Verrillo, 1980; Todd,
1992). Such a function may possibly be used as a heuristic for producing
musically acceptable performances. Of course, local tempo depends on the

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

events of the rhythm and cannot be perceived independently of them
(Desain & Honing, 1991; Gibson, 1975).
In musical performance, variations in local tempo may be either percep-
tible or imperceptible. The delay of important events (i.e., events bearing
strong metrical or structural accents) is a case in point. The existence of
imperceptible delays was demonstrated by Drake, Botte, and Gérard
(1989), who presented an isochronous sequence followed by a longer
event and found that considerable delays in the onset of the longer event
(up to 5% of the IOIs of the preceding events) could not be detected.
Another example of imperceptible variations is the softening of durational
contrasts in performances of alternating IOIs in a nominal 2:1 ratio. In
general, performance timing deviates systematically from metronomic tim-
ing, even when performers are asked to play mechanically or without
expression (Clarke, 1985; Seashore, 1938) - although "mechanical" per-
formances are in some cases preferred to performances by live musicians
(Wapnick & Rosenquist, 1991).
When tempo variations are perceptible, they are called rubato. Rubato
generally involves a gradual or continuous variation in tempo (Robert,
1981). It can function both as a vehicle of musically expression and as a
clarifier of musical structure (Clarke, 1982, 1985; Gabrielsson et al.,
1983; Palmer, 1989; Shaffer, 1981; Sloboda, 1983).
A slowing of local tempo in the vicinity of accents could explain why
2:1 ratios tend tp be softened in music performance whereas 3:1 ratios
(dotted rhythms) are sharpened (as noted above). In the 2:1 case, both
events have metrical accent, the greater accent falling on the onset of the
longer IOI. A slowing of local tempo in the vicinity of the greater accent
would cause the IOI ratio to be reduced (Parncutt, 1987). The ratio may
be formulated (2 + ô):(l+e), where both 8 and e are small positive real
numbers. When e = 5, the ratio is less than two. Musical experience
suggests that the magnitude of the delay of important events is usually
greater than the magnitude of their lengthening, so that e > 5, causing
even more softening of the 2:1 ratio.
In the 3:1 case, the event initiating the shorter IOI does not belong to a
pulse sensation and so has no metrical accent (provided no other IOI ratios
 Pulse Salience and Metrical Accent 449

such as 3:1:2:2 occur in the same context). The temporal position of the
unaccented event is determined only by temporal proximity to its accented
neighbor. Musical experience suggests that the perceived relationship be-
tween the two events may be strengthened simply by reducing this dis-
tance. An interesting analogy in the area of musical intonation is the
sharpening of the leading tone in performance (Nickerson, 1949; Ward,
1970). Apparently, sharpening the leading tone strengthens its melodic or
voice-leading relationship with the tonic scale degree, thereby strengthen-
ing the overall (melodic and harmonic) relationship between tonic and

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

dominant harmonies (Parncutt, 1989). Similarly, sharpening the 3:1 IOI
ratio may yield a "stronger" rhythmic feel. This effect may underlie the
baroque performance convention whereby rhythms notated as dotted
(nominally 3:1) are often performed as if they were double-dotted (nomi-
nally 7:1).
Expressive timing in music performance depends on internal representa-
tions of structure (Clarke, 1985, 1988, 1989; Palmer, 1989; Povel, 1984;
Todd, 1985, 1992). However, analytical representations of rhythmic struc-
ture have so far failed to produce consistently good artificial performances
of a reasonable range of pieces. Rosenthal (1992) has suggested that a
successful algorithmic model of pulse sensation and perceived meter may
aid in the creation of more successful structural representations of rhythm,
and hence improve models of timing. For example, algorithmic formula-
tions of phenomenal and metrical accent may be used to model the tempo-
rary slowing of local tempo in the vicinity of perceptually important
events, as described above. If this approach is successful, the formulations
could then be incorporated into a set of rules for the automatic, yet musi-
cally satisfying, performance of rhythms, along the lines of the perfor-
mances rules described by Sundberg (1988).

TEMPORAL PRECISION AND DISCRIMINATION

hypothesis: The precision with which an event can be timed, and the
precision with which a deviation in the timing of an event can be detected,
depend on the metrical accent of that event.
Two pieces of evidence support this hypothesis. First, experimental data
on performance timing suggest that the timing of metrically stronger beats
is more precise (i.e., less variable) than the timing of metrically weaker
beats. For example, the data of Shaffer (1981) and Shaffer et al. (1985)
imply that beats belonging to the underlying tactus are timed more accu-
rately than other beats. The second piece of evidence concerns the exis-
tence region of pulse sensation. This region corresponds quite well with
the region of greatest sensitivity in synchronization and durational dis-
crimination tasks. Fraisse (1967) presented listeners with two interpolated
450 Richard Parncutt

pulse trains, each of which was isochronous, and asked them to adjust the
time interval between the two sequences to produce isochrony (at double
the tempo of each separate sequence). In another experiment, listeners
were asked whether the sequence was synchronous or not. Performance
was optimal for periods in the vicinity of 500-600 ms (typical range:
200-900 ms), with discrimination thresholds in the vicinity of 4-5%
(typical range: 2-8%). Similar experiments on duration discrimination
and isochrony were performed by Eisler (1975) and Michon (1964).
Fraisse (1982) concluded that synchronization is possible only within the

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

range 200-1800 ms, corresponding to the boundaries of the existence
region of pulse sensation.

PRIMACY, RECENCY, AND THE PSYCHOLOGICAL PRESENT

Real musical rhythms are not usually cyclic, but change as they unfold in
time. Consequently, the relative saliences of the various pulse sensations
evoked by a musical rhythm also vary with time. The model described above
is limited to cyclically repeating rhythms and so does not account for such
variations. The model may be extended to cover ordinary, noncyclic
rhythms by accounting for primacy and recency effects, according to which
the events near the start of a rhythm and near the observation time have a
greater influence on the salience of pulse sensations (and hence on perceived
meter) than intervening events. Both primacy and recency effects are as-
sumed here to involve the duration of the psychological present.

assumption: The perception of pulse is confined to a limited time span

known as the psychological present.
Lerdahl and Jackendoff (1983, p. 21) pointed out that "Metrical struc-
ture is a relatively local phenomenon." The period in time over which
metrical structure exists is called the psychological present (Fraisse, 1963),
otherwise known as the span of consciousness (Wundt, 1874), or specious
present (James, 1950). The psychological (or perceptual) present may be
regarded as a continuous time interval comprising all real-time percepts
and sensations simultaneously available to attention, perception, and cog-
nitive processing. It enables a temporal sequence to be processed as a unit.
It may be regarded as a nonverbal short-term memory possessing a high
degree of structure and organization (Deutsch, 1975). Its duration may be
measured in various ways, for example by testing the ability of listeners to
reproduce sound sequences of various durations. A typical estimate of the
duration of the psychological present is 4 s, with experimental results
ranging from about 2-8 s (Fraisse, 1963; Michon, 1978). Fricke (1989)
proposed a period of 7-15 s. It should be stressed that "because the
present is so highly adaptive, no fixed parameter values can be expected to
 Pulse Salience and Metrical Accent 45 1

describe it adequately" (Michon, 1978, p. 89). The duration of the psycho-

logical present also depends on the time intervals between successive
percepts- the faster the presentation rate, the shorter the memory span.
The maximum number of successive sounds that can be perceived as a unit
has been estimated at 25 (Dietze, 1885; Fraisse, 1982).
In the perception of language, the psychological present enables the
different parts of a sentence or clause to be related to each other and hence
understood (Fraisse, 1982; Wallin, 1901). In music, the psychological
present enables events in a phrase to be related to each other. It appears

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

that both serial and periodic grouping are confined to the psychological
present.

assumption: Recent events in a rhythm have more influence on the sa-

lience of pulse sensations than earlier events.
The effect of the psychological present on the temporal development of
pulse salience may be modeled simply by giving more weight to more
recent events, a procedure that is consistent with the enhanced recall of
recent auditory events in rhythmic sequences (as found, for example, by
Glenberg & Swanson, 1986). In a first approximation, all phenomenal
accents may simply be multiplied by a factor of expj-^/^J, where tx is the
time elapsed between a given event and the observation time, and t/^is the
effective duration of the psychological present in the case of the recency
effect.

assumption: Events near the start of a sequence have a greater influence

on the salience of pulse sensations than later events.
The perceived meter of a sequence is generally established quite early.
According to Lee (1991), "listeners initially assume that the downbeat
falls on the first note and the 'second beat' on the second note, and are
reluctant to revise their assumptions" (p. 98). Longuet-Higgins and Lee
(1982) noted that early events "allow the establishment of metrical hy-
potheses which can then be challenged but not necessarily overthrown by
later evidence" (p. 118) (see also Longuet-Higgins & Steedman, 1971;
Steedman, 1977). In other words, pulse sensations and perceived meter are
established quite quickly and are subsequently quite resilient in the face of
potentially disruptive syncopations. Only when the degree or strength of
syncopation passes a certain threshold will the perceived meter change.
The duration of the primacy effect would appear to be of the same order
of magnitude as the duration of the psychological present. In a first approxi-
mation, the effect may be modeled by multiplying phenomenal accents
• where k is a measure of
by an expression of the form 1 + k exp{- V<Ao}>
the extent of the primacy effect, t0 is the time elapsed between the start of
the sequence and the event, and </f0is a measure of the duration of the
452 Richard Parncutt

psychological present in the case of the primacy effect. Note that ^ and if/0
are not necessarily equal. They will each need to be determined experimen-
tally and can be expected to vary as a function of the perceived rate of the
music.

A DEFINITION OF MUSICAL RHYTHM

The word rhythm may be used to describe any form of temporal period-
icity observable in the physical universe, from molecular vibrations (pe-

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

riod: about 10~13 s) right up to the "rhythm" of expansions and contrac-
tions of the universe between hypothetical big bangs (period: about 1018s).
By comparison to this inconceivably broad spectrum, the periodicities of
musical rhythms are confined to a very narrow range- roughly 200-1800
ms.
Musical rhythm has been defined in various ways. Plato defined rhythm
as order in movement - a definition adopted by Fraisse (1982). Aristox-
enes defined rhythm as organization of time spans (Eggebrecht, 1967).
Martin (1972) defined rhythm as "temporal patterning" (p. 487). Scholes
(1970) gave a more general definition of rhythm: "everything pertaining
to what may be called the time side of music." The Riemann Musik-
lexikon (Eggebrecht, 1967) was more specific, referringto integrated tem-
poral principles of organization and performance in dance, music, and
poetry and emphasizing the dichotomy between the foreground sequence
of rhythmic elements (serial grouping) and the constant underlying move-
ment (periodic grouping). According to Shaffer et al. (1985, p. 63), "Musi-
cal rhythm arises as the interaction of several factors, the time values of
notes and rests forming the melodies, the patterning of metrical accents,
the hierarchy of melodic groupings and the harmonic movement within
these groupings."
A problem with these definitions is that they do not clearly distinguish
musical rhythm from nonrhythm in specific cases. Most definitions refer
to temporal organization, but temporal organization- at least in the form
of serial grouping- is common to both speech and music.
Martin (1972), Jones (1976), and Desain (1992) stressed the impor-
tance of expectancies in rhythm perception; however, expectancies also
occur in nonrhythmic sequences. For example, a particular word may be
expected at a particular temporal position in a sentence. This is not a
rhythmic effect in the musical sense, even though the expected temporal
position may be specified quite precisely. In this light, it is inappropriate to
define musical rhythm in terms of expectancy.
Musical rhythms often include IOIs that approximate simple ratios of
small integers. However Povel and Essens (1985) showed that sequences
composed entirely from IOIs in simple integer ratios (e.g. 1:2 or 1:3) do
 Pulse Salience and Metrical Accent 453

not necessarily evoke pulse sensations (internal clocks) and so are not
necessarily rhythmic in the musical sense. In any case, a definition of
rhythm based on approximations to simple ratios of small integers would
be plagued by the arbitrariness of the concepts of approximation and
simplicity.
Periodic perceptual grouping appears to represent the most appropriate
basis for a definition of musical rhythm. Periodic grouping is typical in
music, but relatively unusual in speech- with the possible exception of
mechanical renditions of poetry. Periodic grouping, or pulse sensation, is

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

thus capable of distinguishing musical rhythm from the rhythm of regular
speech. Consider the following concise definition.

definition: A musical rhythm is an acoustic sequence evoking a sensation

of pulse.
This definition is reminiscent of the generally accepted definition of a
tone as a sound that evokes a sensation of pitch (American Standards
Association, 1960). Both definitions are clear and simple, and they corre-
spond closely to the meanings of the terms tone and rhythm in everyday
language.
A sensation of pulse may be regarded as a trivial form of expectancy.
Once a pulse sensation is established, events are "expected" at equal time
intervals. A definition of rhythm based on pulse sensation thus automati-
cally incorporates the notion of expectancy.
The existence of pulse sensations in musical rhythms may be demon-
strated by analyzing the timing of rhythmic production or musical perfor-
mance. Inspired by the work of Lashley (1951), Martin (1972) defined
rhythm as "relative timing between adjacent and non-adjacent elements in
a behavior sequence" (abstract), implying that the temporal position of
each sound event in a rhythmic sequence may be determined by all other
events in the psychological present- not just immediately preceding and
following events. In an analysis of performances of Algerian and Carib-
bean music, Kluge (1990) found that adjacent IOIs within a phrase were
negatively correlated- if one was lengthened, its neighbor was shortened,
and vice-versa- however, this was not the case between phrases. These
results are consistent with the existence of pulse sensations within limited
time spans. The negative correlation between timing deviations of adjacent
events suggests that a strong pulse feel was maintained by monitoring the
timing of nonadjacent events.

hypothesis: The greater the (aggregate) salience of the pulse sensation(s)

evoked by a sequence, the greater the sequence's overall rhythmicity.
Povel (1984, p. 317) considered that "the most outstanding characteris-
tic of a rhythm is what may be called its rhythmical value [i.e., its rhyth-
454 Richard Parncutt

micity],which tentativelycan be describedas the amountof tension that

accompaniesthe perceptionof a rhythm."Experimentalresultsof Povel
(1981), Essensand Povel (1985), Summers,Hawkins,and Mayers(1986),
and Drake and Gérard(1989) furthersuggestthat the greaterthe rhyth-
micity,the easierit is to understand(perceive,process,encode)and subse-
quently to reproducea temporalsequence.In short, rhythmicsequences
(i.e., sequencesimplying a metricalframework)are easier to reproduce
than nonrhythmicsequences.

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

Conclusions

The resultsof this experimentalpartof the studymaybe summarizedas

follows:
1. The underlying beat (tactus) of a cyclically repeating se-
quence of identical sounds was generallyquite ambiguous.
Eachgiven rhythmevokeda numberof differentpulse sensa-
tions. The position of the downbeatin each cycle was simi-
larly ambiguous.
2. The relativesalienceof the differentpulse sensationsevoked
by a rhythmdependedto a considerableextenton both rhyth-
mic pattern (the sequenceof IOI ratios) and note rate (the
numberof notes per unit time).
3. Rhythmicambiguitytendedto increasewith increasingrhyth-
mic complexity.
4. The likelihood of an event belongingto the tactus (or act-
ing as a downbeat) increasedwith the IOI following the
event, implying that events followed by longer IOIs were
perceivedas more accented.However, this effect was only
consistently observed at faster tempi, that is, among rela-
tively short IOIs. At slower tempi, the reverse effect was
sometimesobserved.
5. The tempiof most pulseresponseslay in the vicinityof moder-
ate tempo (approx.600-700 ms). The tactustendedto gravi-
tate toward this tempo regardlessof the actual note rate of
the rhythm.
6. Subjectivegroupingin isochronoussequencesoccurredmore
commonlyby foursthan by threesover a wide rangeof musi-
cal templ.
The model developedand testedin this paperwas basedon the follow-
ing assumptions:
 Pulse Salience and Metrical Accent 455

1. Durational accent increases with the IOI that follows an

event, but saturates as that IOI approaches about Is, corre-
sponding to the duration of the echoic store.
2. Durational accent approaches zero as IOI approaches the
minimum discriminate IOI (typically, 50-100 ms).
3. The perception of pulse in musical rhythms may be regarded
as a form of pattern recognition.
4. The salience of a pulse sensation increases with the phenome-
nal accent and number of events matching that pulse.

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

5. The salience of pulse sensations is greatest in the vicinity of
100 events per minute (600 ms) and falls to zero for tempi
outside the approximate range 33-300 events per minute
(200-1800 ms).
6. The effect of tempo on pulse salience is largely independent
of the effect of matching events.
7. The metrical accent of a rhythmic event depends on the sa-
liences of the pulse sensations including that event.
A literature survey suggested the following additional hypotheses:
1. The salience of a perceived meter may be estimated by sum-
ming the saliences of the consonant pulse sensations of which
it is composed.
2. The salience of a pulse sensation may be enhanced by
other, consonant pulse sensations. Enhancement of pulse sa-
lience may explain certain perceptual limitations on beat
subdivision.
3. Temporal categories in rhythm perception are determined by
metrical accents.
4. Temporal precision and discrimination are enhanced by metri-
cal accentuation.
5. Expressive timing may be modeled by a slowing of local
tempo near metrical and structural accents. The degree of
slowing depends on the strength of the accents.
6. Variations in pulse and meter salience as a function of time in
real, noncyclic rhythms may be modeled by assuming that the
perception of pulse is confined to the psychological present
(recency effect) and by assigning extra weight to events near
the start of the sequence (primacy effect).
7. A musical rhythm may be defined as a temporal sequence
evoking a sensation of pulse.
The present model is nonhierarchical, assuming the concurrent and
largely independent existence of pulse sensations of different periods and
456 Richard Parncutt

saliences. According to Monahan and Hirsch (1990), hierarchicalrepresen-

tations of meter are unnecessary for explanations of temporal discrimina-
tion of events in simple rhythms. More generally, Swain (1986) has
pointed out that any hierarchical theory of music perception is subject to
known limitations in the amount of information that humans are able to
accept and process. The success of the present model in accounting for the
results of the above experiments suggests that hierarchical representations
of meter may also be unnecessary for an explanation of pulse and accent
strength, at least in relatively simple rhythms.1

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

References
Agawu, V. K. The rhythmic structure of West African music. Journal of Musicology, 1987,
5, 400-418.
Allan, L. G. The perception of time. Perception & Psychophysics, 1979, 26, 340-354.
American Standards Association. USA standard acoustical terminology. New York: Ameri-
can Standards Association, 1960.
Apel, W. Harvard dictionary of music, 2nd ed. Cambridge, MA: Harvard University Press,
1970.
Bamberger,J. Intuitive and formal musical knowing: Parablesof cognitive dissonance. In S.
S. Madeja (Ed.), The arts, cognition and basic skills. New Brunswick, NJ: Transactions,
1978.
Bamberger,J. Cognitive structuring in the apprehension and description of simple rhythms.
Archives of Psychology, 1980, 48, 177-199.
Bamberger, J. Revisiting children's drawings of simple rhythms: A function for reflection-
in-action. In S. Strauss (Ed.), U-shaped behavioral growth. New York: Academic Press,
1982.
Beauvillain, C, & Fraisse, P. On the temporal control of polyrhythmic performance. Music
Perception. 1984. 1. 485-499.
Bengtsson, I. Notation, motion and perception: Some aspects of musical rhythm. In A.
Gabrielsson (Ed.), Action and perception in rhythm and music. Stockholm: Royal Swed-
ish Academy of Music, 1987, pp. 69-80.
Bengtsson, I., & Gabrielsson, A. Analysis and synthesis of musical rhythm. In J. Sundberg
(Ed.), Studies of music performance. Stockholm: Royal Swedish Academy of Music,
1983.
Benjamin, W. A theory of musical meter. Music Perception, 1984 1, 355-413.
Bolton. T. L. Rhvthm. American Journalof Psvcholoev. 1894. 6. 145-238.
Braun, F. Untersuchungen iiber das persônliche Tempo [Investigations into personal
tempo]. Archiv fur die gesamte Psychologie, 1927, 60, 317-360.
Bregman, A. S., & Pinker, S. Auditory streaming and the building of timbre. Canadian
Journal of Psychology, 1978,32, 19-31.

1. I am grateful to Johan Sundberg, Erik Jansson, Anders Askenfelt, Sten Ternstrom,

and Gunilla Carlsson for providing a friendly working environment, discussions, help,
advice, and computer facilities. Alf Gabrielsson advised on the design and running of the
experiments. For valuable comments on earlier drafts of the manuscript, I thank Jamshed
Bharucha, Chris Childs, Annabel Cohen, Carolyn Drake, Bradley Frankland, Gunter
Kreutz, Justin London, David Rosenthal, and an anonymous reviewer. The research was
supported by fellowships from the Swedish Institute and from the Natural Sciences and
Engineering Research Council of Canada.
 Pulse Salience and Metrical Accent 457

Brown, J. C. Determination of the meter of musical scores by autocorrelation. Journal of

the Acoustical Society of America, 1993,94, 1953-1957.
Brown, W. Temporal and accentual rhythm. Psychological Review, 1911, 18, 336-346.
Buytendijk, F. J. J., oc Meesters, A. Duration and course or the auditory sensation. Com-
mentationes Pontificia Academia Scientiarum, 1942, 6, 557-576.
Clarke, E. F. Timing in the performance of Erik Satie's 'Vexations'. Acta Psychologica,
1982,50, 1-19.
Clarke, E. F. Some aspects of rhythm and expression in performances or Erik Satie s
"Gnossienne No. 5." Music Perception, 1985, 2, 299-328.
Clarke, E. F. Categorical rhythm perception: An ecological perspective. In A. Gabnelsson
(Ed.), Action and perception in rhythm and music. Stockholm: Royal Swedish Academy
of Music, 1987, pp. 19-34.

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

Clarke, E. F. Generative principles in music performance. In J. Sloboda (Ed.), Generative
processes in music: The psychology of performance, improvisation and composition.
Oxford: Science Publications, 1988.
Clarke, E. F. The perception of expressive timing in music. Psychological Research, 1989,
51, 2-9.
Collyer, C. E., Broadbent, H. A., & Church, R. M. Categorical time production: Evidence
for discrete timing in motor control. Perception and Psychophysics, 1992, 51, 134-144.
Cooper, G., & Meyer, L. B. The rhythmic structure of music. Chicago: University or
Chicago Press, 1960.
Crowder, R. G. Improved recall for digits with delayed recall cues, journal of Expérimental
Psychology,1969, 82, 258-262.
Crowder, R. G. The role of one's own voice in immediate memory. Cognitive Psychology,
1990, 1, 157-178.
Crowder, R. G., & Morton, J. Precategorical acoustic storage (P.A.S.). Perception & Psy-
chophysics, 1969, 5, 365-373.
Desain, P. A (de)composable theory of rhythm perception. Music Perception, lyyl, y,
439-454.
Desain, P., & Honing, H. Quantization or musical time: A connectionist approacn. com-
puter Music Journal, 1992, 13(3), 56-66.
Desain, P., & Honing, H. Tempo curves considered harmful. Proceedings of the 1V91
International Computer Music Conference. San Francisco: Computer Music Associa-
tion, 1991, pp. 143-149.
Deutsch, D. The organization of short-term memory tor a single acoustic attribute. In U.
Deutsch & J. A. Deutsch (Eds.), Short-term memory. New York: Academic Press, 1975,
pp. 108-151.
Deutsch, D. Grouping mechanisms in music. In D. Deutsch (Ed.), The psychology of music.
New York: Academic Press, 1982.
Dietze, G. Untersuchungen iiber den Umfang des Bewufêteins bei regelmâisig aufeinan-
derfolgenden Schalleindrucken [Investigations into the span of consciousness for regular
successions of sound impressions]. Philosophische Studien, 1885, 2, 362-393.
Drake, C. Reproduction of musical rhythms by children, adult musicians, and adult
nonmusicians. Perception & Psychophysics 1993, 53, 25-33.
Drake, C, & Botte, M-C. Tempo sensitivity in auditory sequences: Evidence for a multiple-
look model. Perception & Psychophysics (in press).
Drake, C, botte, Jvi-^., oc uerara, ^. /\ perceptual aisturuun m simyie rnusu,uirnywrn*.
Paper presented at the International Conference on Psychophysics, Cassis, France,
1989.
Drake, C, Dowling, J., & Palmer, C. Accent structures in the reproduction or simple tunes
by children and adult pianists. Music Perception, 1991, 8(3), 315-334.
Drake, C, & Gérard, C. A psychological pulse train: How young children use this cogni-
tive framework to structure simple rhythms. Psychological Research, 1989, 51, 16-22.
Duke, R. A., Geringer,J. M., & Madsen, C. K. Performance ot perceived beat in relation to
age and music training. Journal of Research in Music Education, 1991, 39(1), 35-45.
458 Richard Parncutt

Eggebrecht, H. H. Riemann Musiklexikon. Mainz: Schott, 1967.

Ehrlich, S. Le mécanisme de la synchronisation sensori-motorice. Année Psychologique,
1958, 58. 7-23.
Eisler, H. Subjective duration and psychophysics. Psychological Review, 1975, 82, 429-
450.
Essens. P. J., ÔcPovel, D. J. Metrical and nonmetrical representation of temporal patterns.
Perception & Psychophysics, 1985, 37, 1-7.
Fowler, C. A. "Perceptual centers" in speech production and perception. Perception &
Psychophysics, 1979, 25, 375-388.
Fraisse, P. Les structures rythmiques. Louvain: Publications Universitaires de Louvam,
1956.
Fraisse, P. The psychology of time. New York: Harper & Row, 1963.

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

Fraisse, P. Le seuil différentiel de durée dans une suite régulière d'intervalles [The differen-
tial duration threshold in a regular series of intervals]. Année Psychologique, 1967, 1,
43-49.
Fraisse, P. Les synchronisations sensori-motorices aux rythmes. In J. Requin (Ed.), Anticipa-
tion et comportement. Paris: Editions CNRS, 1980.
Fraisse P. Rhythm and tempo. In D. Deutsch (Ed.), The psychology of music. New York:
Academic Press, 1982, pp. 149-180.
Fraisse, P. A historical approach to rhythm as perception. In A. Gabrielsson (Ed.), Action
and perception in rhythm and music. Stockholm: Royal Swedish Academy of Music,
1987, dd. 7-18.
Fraisse, P., Pichot, P., & Clairouin, G. Les aptitudes rythmiques: Etude comparée des
oligophrènes et des enfants normaux. Journal de Psychologie Normale et Pathologique,
1949.42,309-330.
Friberg, A., Frydén, L., Bodin, L., & Sundberg,J. Performancerules for computer-controlled
contemporary keyboard music. Computer Music Journal, 1991, 15(2), 49-55.
Fncke, J. P. Merkmale und Bedingungen des Sprachhchen in der Musik [Characteristics
and conditions of linguistic aspects of music]. In J. P. Fricke (Ed.), Die Sprache der
Musik. Regensburg: Bosse, 1989, pp. 171-188.
Fnscheisen-Kohler, I. Feststellung des weder langsamen noch schnellen (mittelmaisigen)
Tempos [Determination of moderate (neither slow nor fast) tempo]. Psychologische
Forschung, 1933, 18, 291-298.
Gabrielsson, A. Performance of rhythm patterns. Scandinavian Journal of Psychology,
1974, 15, 63-72.
Gabrielsson, A., Bengtsson, I., & Gabrielsson, B. Performance of musical rhythm in 3/4
and 6/8 meter. Scandinavian Journal of Psychology, 24, 193-213.
Garner, W. R. The processing of information and structure. New York: Wiley, 1974.
Garner, W. R., & Gottwald, R. L. The perception and learning of perceptual patterns.
Quarterly Journal of Experimental Psychology, 1968, 20, 97-109.
Gérard, C., & Drake, C. The inability of young children to reproduce intensity differences
in musical rhythms. Perception & Psychophysics, 1990, 48, 91-101.
Gibson, J. J. Events are perceivable, but time is not. In J. T. Fraser & N. Lawrence (Eds.),
The study of time, Vol. 2. Berlin: Springer-Verlag, 1975.
Gibson, J. J. the ecological approach to visual perception. Boston: Houghton Mifflin,
1979.
Glenberg, A. M., & Jona, M. Temporal coding in rhythm tasks revealed by modality
effects. Memory & Cognition, 1991, 19, 514-522.
Glenberg, A. M., & Swanson, N. C. A temporal distinctiveness theory of recency and
modality effects. Journal of Experimental Psychology: Learning, Memory, & Cogni-
tion, 1986, 12, 3-15.
Glenberg, A. M., Mann, S., Alman, L., Forman, T., & Procise, S. Modality effects in the
coding of rhythms. Memory & Cognition, 1989, 17, 373-383.
Glucksberg, S., ÔCCowen, G. N., Jr. Memory for non-attended auditory material. Cogni-
tive Psychology,1970, 1, 149-156.
 Pulse Salience and Metrical Accent 459

Grant, R. E., & LeCroy, S. Effects of sensory mode input on the performance of rhythmic
perception tasks by mentally retarded subjects, journal of Music Therapy, 1986, 23,
2-9.
Guttman, N., & Julesz, B. Lower limits of auditory periodicity analysis, journal of the
Acoustical Society of America. 1963. 35. 610.
Guttmann, A. Das Tempo und seine Variationsbreite [Tempo and its range of variation].
Zeitschrift fur angewandte Psychologie, 1931, 40, 65.
Handel, S. Perceiving melodic and rhythmic auditory patterns. Journal of Experimental
Psychology, 1974, 103, 922-933.
Handel, S. Tempo in rhythm: Comments on Sidnell. Psychomusicology, 1986, 6, 19-23.
Handel, S., & Buffardi, L. Pattern perception: Integrating information presented in two
modalities. Science, 1968, 162, 1026-1028.

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

Handel, S., ôc Lawson, G. R. The contextual nature of rhythmic interpretation. Perception
& Psychophysics, 1983,34, 103-120.
Handel, S., & Oshinsky, J. S. The meter of syncopated auditory polyrhythms. Perception &
Psychophysics, 1981,30, 1-9.
Harrell, T. W. Factors influencing preference and memory for auditory rhythms. Journal of
General Psychology, 1937, 16, 427-469.
Harrison, R. Personal tempo and the interrelationship or voluntary and maximal rates or
movement. Journal of General Psychology, 1941, 24, 343-379.
Hlawicka, K. Die rhythmische Verwechslung [Rhythmic ambiguity]. Musikforschung,
1958,11,33-49.
Howell, P. Prediction of P-center location from the distribution of energy in the amplitude
envelope: I. Perception & Psychophysics, 1988, 43, 90-93.
James, W. Principles of psychology. New York: Dover, 1950.
Jones, A. M. Studies in African music. London: uxrora university rress, iy$y.
Jones, M. R. Time, our lost dimension: Toward a new theory of perception, attention, and
memory. Psychological Review, 1976, 83, 323-355.
Jones, M. R. Dynamic pattern structure in music: Recent theory and research. Perception
& Psychophysics, 1987a, 41, 621-634.
Jones, M. R. Perspectives on musical time. In A. Gabrielsson (Ed.), Action and percep-
tion in rhythm and music. Stockholm: Royal Swedish Academy of Music, 1987b, pp.
153-176.
Jones, M. R., & Boltz, M. Dynamic attending and responses to time. Psychological Review,
1989, 96, 459-491.
Jones, M. R., Boltz, M., & Kidd, G. Controlled attending as a function of melodic and
temporal context. Perception & Psychophysics, 1982, 32, 211-218.
Jones, M. R., Kidd, G., & Wetzel, R. Evidence for rhythmic attention. Journal of Experi-
mental Psychology: Human Perception & Performance, 1981, 7, 1059-1073.
Keele, S. W., Pokony, R. A., Corcos, D. M., & Ivry, R. Do perception and motor produc-
tion share common timing mechanisms: A correlational analysis. Acta Psychologica,
1985.60. 173-191.
Kluge, R. Zeitliche Bezugssysteme als Basis rhythmischer Stile [Temporal reference frames
as a basis for rhythmic styles]. In H. Braun (Ed.), Problème der Volksmusikforschung,
1990, pp. 47-60.
Koetting, J. What do we know about African rhythm? Ethnomusicology, 1986, 30, 6-31.
Krebs, H. Some extensions or the concepts or metrical consonance and dissonance, journal
of Music Theory, 1987, 31, 99-120.
Kronman, U., & Sundberg, J. Is the musical retard an allusion to physical motion? In A.
Gabrielsson (Ed.), Action and perception in rhythm and music. Stockholm: Royal Swed-
ish Academy of Music, 1987, pp. 57-68.
Kubovy, M., & Howard, F. P. Persistence of pitch-segregating echoic memory. Journal of
Experimental Psychology: Human Perception & Performance, 1976, 2, 531-537.
Lashley, K. S. The problem of serial order in behavior. In L. A. Jettnes (hd.), Cerebral
mechanisms in behavior. New York: Wiley, 1951.
460 Richard Parncutt

Lee, C. S. The perception of metrical structure: Experimental evidence and a model. In P.

Howell, R. West, & I. Cross (Eds.), Representing musical structure. London: Academic
Press, 1991, pp. 59-127.
Lerdahl, F., & Jackendoff, R. On the theory of grouping and meter. Musical Quarterly,
1981.25.45-90.
Lerdahl, F., & Jackendoff, R. A generative theory of tonal music. Cambridge, MA: MIT
Press, 1983.
Longuet-Higgins, H. C. The perception of melodies. Nature (London), 1976, 263, 646-653.
Longuet-Higgins, H. C, & Lee, C. S. The perception of musical rhythms. Perception,
1982,11, 115-128.
Longuet-Higgins, H. C, & Lee, C. S. The rhythmic interpretation of monophonie music.
MusicPerception,1984, 1, 424-441.

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

Longuet-Higgins, H. C, & Steedman, M. J. On interpreting Bach. In B. Meltzer & D.
Michie (Eds.), Machine intelligence, Vol. 6. Edinburgh: Edinburgh University Press,
1971.
MacDougall, R. The structure of simple rhythm forms. Psychological Review, Monograph
Supplements, 1903, 4, 309-416.
Mach, E. Die Analyse der Empfindungen [Analysis of sensations], 1st ed. Jena: Fischer,
1886. (2nd éd., 1911)
Marcus, S. M. Acoustic determinants of perceptual (P-center) location. Perception & lJsy-
chophysics,1981, 30, 247-256.
Martin, J. G. Rhythmic (hierarchical) versus serial structure in speech and other behavior.
PsychologicalReview,1972, 79, 487-509.
Massaro, D. W. Preperceptualauditory images. Journal of Experimental Psychology, 1970,
55,411-417.
Massaro, D. W. Perceptual units in speech recognition. Journal of Experimental Psychol-
ogy. 1974, 102, 199-208.
Meumann, E. Untersuchungen zur Psychologie und Aesthetik des Rhythmus [Investiga-
tions into the psychology and aesthetics of rhythm]. Philosophische Studien, 1894, 10,
249-322, 393-430.
Michon, J. A. Studies on subjective duration. 1. Differential sensitivity in the perception of
repeated temporal intervals. Acta Psychologica, 1964, 22, 441-450.
Michon, J. A. Timing in temporal tracking. Soesterberg, NL: Institute for Perception TNO,
1967.
Michon, J. A. Programs and "programs" for sequential patterns in motor behaviour, brain
Research, 1974, 71, 413-424.
Michon, J. A. Time experience and memory processes. In J. T. Fraser ôc N. Lawrence
(Eds.), The study of time, Vol. 2. Berlin: Springer-Verlag, 1975.
Michon, J. A. The making of the present: A tutorial review. In J. Requin (bd.), Attention
and performance, Vol. 7. New York: Academic Press, 1978.
Miles, D. W. Preferred rates in rhythmic response, journal of General Psychology, 1^3/,
16, 427-469. _ .
Miller, G. A., & Heise, G. A. The trill threshold. Journal of the Acoustical Society of
America, 1950, 22, 637-638.
Miner, J. B. Motor, visual and applied rhythms. Psychological Keview, Jvionograpnsupple-
ments, 1903, 5, 1-106.
Mishima, J. Introduction to the morphology of human Oehavior: i m expérimentai siuay
of mental tempo. Tokyo: Tokyo Publishing, 1965.
Miyake, I. Researches on rhythmic action. Studies from the YalePsychological Laboratory,
1902. 10. 1-48.
Monahan, C. B., & Hirsch, I. J. Studies in auditory timing: 2. Rhythm patterns. Perception
& Psychophysics,1990, 47, 227-242.
Moore, B. C. J., & Glasberg, B. R. Growth of forward masking for sinusoidal and noise
maskers as a function of signal delay: Implications for suppression in noise. Journal of
the AcousticalSocietyof America,1983, 73, 1249-1259.
 Pulse Salience and Metrical Accent 461

Morton, J., Marcus, S., & Frankish, C. Perceptual centers (P-centers). Psychological Re-
view, 1976, 83, 405-408.
Neisser, U. Cognitive psychology. New York: Appleton-Century-Crofts, 1967.
Nickerson, J. h Intonation or solo and ensemble performance or the same melody. Journal
of the Acoustical Society of America, 1949, 21, 593-595.
Noorden, L. van. Temporal coherence in the perception of tone sequences. Unpublished
doctoral dissertation, Institute for Perception Research, Eindhoven, The Netherlands,
1975.
Palmer, C. Mapping musical thought to musical performance. Journal of Experimental
Psychology: Human Perception & Performance, 1989, 15, 331-346.
Palmer, C, & Krumhansl, C. L. Mental representations for musical meter. Journal of
Experimental Psychology: Human Perception & Performance, 1990, 16, 728-741.

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

Parncutt, R. Algorithm for event, pulse and metre salience in musical rhythms. Poster
presented at the International Conference on Event Perception and Action, Uppsala,
Sweden, 1985.
Parncutt, R. The perception of pulse in musical rhythm. In A. Gabrielsson (Ed.), Action
and perception in rhythm and music. Stockholm: Royal Swedish Academy of Music,
1987, pp. 127-138.
Parncutt, R. Harmony: A psychoacoustical approach, Berlin: bpnnger-Verlag, 1?»?.
Patterson, W. M. The rhythm of prose. New York, 1916.
Plomp, R. Rate of decay of auditory sensation. Journal of the Acoustical Society of Amer-
ica, 1964, 36, 277-282.
Plomp, R. Pitch of complex tones. Journal of the Acoustical Society of America, 1967, 41,
1526-1533.
Plomp, R., & Bouman, M. A. Relation between hearing threshold and duration for tone
pulses. Journal of the Acoustical Society of America, 1959, 31, 749-758.
Povel, D. J. Internal representation of simple temporal patterns. Journal of Experimental
Psychology: Human Perception & Performance, 1981, 7, 3-18.
Povel, D. J. A theoretical framework for rhythm perception. Psychological Research, 1984,
45, 315-337.
Povel, D. J., & Essens, P. Perception of temporal patterns. Music Perception, 1985, 2,
411-440.
Povel, D. J., & Okkerman, H. Accents in equitone sequences. Perception & Psychophysics,
1981.30.565-572.
Repp, B. H. Patterns of expressive timing in performances of a Beethoven minuet by
nineteen famous pianists. Journal of the Acoustical Society of America, 1990, 88,
622-641.
Reymert, M. L. The personal equation in motor capacities. Scandinavian Science Kevtew,
1923,2, 177-222.
Riemann, H. Musikalische Dynamik und Agogik. Hamburg, St. Petersburg, 1884.
Rimoldi, H. J. A. Personal tempo. Journal of Abnormal and Social Psychology, 1951, 46,
283-303.
Ritsma, R. J. Frequencies dominant in the perception of the pitch of complex tones. Journal
of the Acoustical Society of America, 1967, 42, 191-198.
Robert, W. Chopin's tempo rubato in theory and practice. Piano Quarterly, 1981, 113,
42-44.
Roederer, J. G. Introduction to the physics and psychophysics of music, 2nd éd., 2nd
reprint, corrected. New York: Springer-Verlag, 1979.
Rosenthal, D. A model of the process of listening to simple rhythms. Music Perception,
1989,6,315-328.
Rosenthal, D. Emulation of human rhythm perception. Computer Music Journal, IWA,
16(1), 64-76.
Rostron, A. B. Brief auditory storage: Some further observations. Acta Psychologica, 1974,
38. 471-482.
Rothstein, W. Phrase rhythm in tonal music. New York: Schirmer, 1989.
462 Richard Parncutt

Rowlands, L. Rhythmic relationships in Ghanaian music and dance. Unpublished M. Litt

thesis, University of New England, Armidale NSW, Australia, 1991.
Royer, F. L., & Garner, W. R. Perceptual organization of nine-element auditory temporal
patterns. Perception & Psychophysics, 1970, 7, 115-120.
Schab, F. R., & Crowder, R. G. Accuracy of temporal coding: Auditory-visual comparison.
Memory & Cognition, 1989, 17, 384-397.
Scholes, P. A. (Ed.). Oxford companion to music. London: Oxford University Press,
1970.
Schulze, H. The detectability of local and global displacements in regular rhythmic pat-
terns. Psychological Research, 1978,40, 173-181.
Schulze, H. H. Categorical perception of rhythmic patterns. Psychological Research, 1989,
51, 10-15.

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

Schiitte, H. Ein Funktionsschema fur die Wahrnehmung eines gleichmâtëigenRhythmus in
Schallimpulsfolgen [A model for the perception of isochrony in sequences of sound
pulses]. Biological Cybernetics, 1978, 29, 49-55.
Seashore, C. E. Motor ability, reaction-time, rhythm, and time-sense. University of Iowa
Studies in Psychology, 1899, 2, 64-84.
Seashore, C. E. Psychology of music. New York: McGraw-Hill, 1938.
Shaffer, L. H. Performances of Chopin, Bach, and Bartok: Studies in motor programming.
Cognitive Psychology, 1981, 13, 326-376.
Shaffer, L. H., Clarke, E. F., ÔcTodd, N. P. Metre and rhythm in piano playing. Cognition,
1985,20,61-77.
Shaffer, L. H., & Todd, N. P. The interpretative component in musical performance. In A.
Gabrielsson (Ed.), Action and perception in rhythm and music. Stockholm: Royal Swed-
ish Academy of Music, 1987, pp. 139-152.
Simon, H. A. Perception du pattern musical par AUDITEUR. Sciences de I*Art, 1968, 5(2),
28-34.
Sloboda, J. The communication of musical metre in piano performance. QuarterlyJournal
of Experimental Psychology, 1983, 35A, 377-396.
Sloboda, J. A. The musical mind: The cognitive psychology of music. Oxford: Clarendon,
1985.
Smith, J. Reproduction and representation of musical rhythms: The effects of musical skill.
In D. R. Rogers & J. A. Sloboda (Eds.), Acquisition of symbolic skill. New York:
Plenum, 1983.
Smith, K. C, & Cuddy, L. L. Effects of metric and harmonic rhythm on the detection of
pitch alternations in melodic sequences. Journal of Experimental Psychology: Human
Perception and Performance, 1989, 15, 457-471.
Squire, C. R. A genetic study of rhythm. American Journal of Psychology, 1901, 12, 492-
589.
Steedman, M. J. The perception of musical rhythm and metre. Perception, 1977, 6, 555-
569.
Stern, W. Das psychische Tempo (Psychological tempo). In Ùber die Psychologie der
individuellen Differenzen. Leipzig: Barth, 1900.
stetson, R. H. A motor theory or rhythm and discrete succession. Psychological Review,
1905, 12, 250-270, 292-350.
Summers, J. J., Bell, R., ÔCBurns, B. D. Perceptual and motor factors in the imitation of
simple temporal patterns. Psychological Research, 1989, 51, 23-27.
Summers, J. J., Hawkins, S. R., 6c Mayers, H. Imitation and production of interval ratios.
Perception & Psychophysics, 1986, 39, 437-444.
Sundberg, J. Computer synthesis of music performance. In J. A. Sloboda (Ed.), Generative
processes in music. Oxford: Clarendon, 1988.
Sundberg, J., Askenfelt, A., & Frydén, L. Musical performance: A synthesis-by-rule ap-
proach. Computer Music Journal, 1983, 7, 37-43.
Sundberg, J., & Verrillo, V. On the anatomy of the retard: A study of timing in music.
Journal of the Acoustical Society of America, 1980, 68, 772-779.
 Pulse Salience and Metrical Accent 463

Swain, D. The need for limits in hierarchical theories of music. Music Perception, 1986, 4,
121-148.
Temperley,N. M. Personal tempo and subjective accentuation. Journal of General Psychol-
ogy, 1963, 68, 267-287.
Terhardt, E., Stoll, G., & Seewann, M. Algorithm for extraction of pitch and pitch salience
from complex tonal signals. Journal of the Acoustical Society of America, 1982, 71,
679-688.
Thomassen, M. T. Melodic accent: experiments and a tentative model. Journal of the
Acoustical Society of America, 1982, 71, 1596-1605.
Thompson, W. E, Sundberg, J., Friberg, A., & Frydén, L. The use of rules for expression in
the performance of melodies. Psychology of Music, 1989, 17, 63-82.
Todd, N. P. McA. A model of expressive timing in tonal music. Music Perception, 1985, 3,

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020

33-58.
Todd, N. P. McA. The dynamics of dynamics: A model of musical expression. Journal of
the Acoustical Societv of America. 1992. 91. 3540-3550.
Treisman, A. M. Monitoring and storage of irrelevant messages in selective attention.
Journal of Verbal Learning & VerbalBehavior, 1964, 3, 449-459.
Treisman, A. M., & Howarth, C. I. Changes in threshold level produced by a signal
preceding or following the threshold stimulus. Quarterly Journal of Experimental Psy-
chology, 1959, 11, 129-142.
Treisman, A. M., & Rostron, A. B. Brief auditory storage: A modification of Sperling's
paradigm applied to audition. Acta Psychologica, 1972, 36, 161-170.
Upitis, R. Children's understanding of rhythm: The relationship between development and
music training. Psychomusicology, 1987, 7, 41-60.
Vorberg, D., & Hambuch, R. On the temporal control of rhythmic performance.
In J. Requin (Ed.), Attention and performance, Vol. 7. New York: Academic Press,
1978.
Vos, J., & Rasch, R. The perceptual onset of musical tones. Perception & Psychophysics,
1981,29,323-335.
Vos, P. G. Pattern perception in metrical tone sequences. Unpublished thesis, University or
Nijmegen, 1973.
Vos, P. G. Temporal duration factors in the perception of auditory rhythmic patterns.
Scientific Aesthetics, 1977, 1, 183-199.
Vos, P. G., Collard, R. F., & Leeuwenberg, E. L. What melody tells about meter in music.
Zeitschrift fur Psychologie, 1981, 189, 25-33.
Vuori, M. Figurai and temporal structures in children's sight-reading performance. Cana-
dian Journal of Research in Music Education, 1991, 33, 201-206.
Wallin, J. E. W. Researches on the rhythm of speech. Studies from the Yale Psychological
Laboratory, 1901, 9, 1-142.
Wallin, J. E. W. Experimental studies of rhythm and time. Psychological Review, 1911, 18,
100-131, 202-222: 1912, 19, 271-298.
Wapnick, J., & Rosenquist, M.-J. Preferences of undergraduate music majors for se-
quenced versus performed piano music. Journal of Research in Music Education, 1991,
39(2), 152-160.
Ward, W. D. Musical perception. In J. V. Tobias (Ed.), Foundations of modern auditory
theory. New York: Academic Press, 1970.
Wessel, D. L. Timbre space as a musical control structure. Computer Music Journal, 1979,
3(2), 45-52.
Wing, A. M. The timing of interresponse intervals. Hamilton, Ontario: Department ot
Psychology, McMaster University, 1973, Technical Report No. 56.
Wing, A. M., & Knstofferson, A. B. The timing or interresponse intervals. Perception cr
Psychophysics, 1973a, 13, 455-460.
Wing, A. M., & Knstofferson, A. B. Response delays and the timing or discrete motor
responses. Perception & Psychophysics, 1973b, 14, 5-12.
Woodrow, H. The role of pitch in rhythm. Psychological Review, 1911, 1», M-/1.
464 Richard Parncutt

Woodrow, H. S. Time perception. In S. S. Stevens (Ed.), Handbook of experimental psy-

chology. New York: Wiley, 1951, pp. 1224-1236.
Wu, C. F. Personal tempo and speed in some rate tests. Psychological Abstracts, 1935, 9,
1709.
Wundt, W. Grundzuge der physiologischen Psychologie [Foundations of physiological
psychology]. Leipzig: Engelmann, 1874.
Yeston, M. The stratification of musical rhythm. New Haven, CT: Yale University Press,
1976.
Zwicker, E., & Fasti, H. Zur Abhàngigkeit der Nachverdeckung von der Stôrimpulsdauer
[On the dependence of forward masking on masker duration]. Acustica, 1972, 26,
78-82.

Downloaded from http://online.ucpress.edu/mp/article-pdf/11/4/409/145282/40285633.pdf by guest on 26 July 2020