HST.

725: Music Perception and Cognition, Spring 2009

Harvard-MIT Division of Health Sciences and Technology
Course Director: Dr. Peter Cariani

Rhythm: patterns
of events in time

Thursday, May 14, 2009 Courtesy of John Hart (http://nanobliss.com). Used with permission.
 1. Rhythmic pattern induction & expectation
chunking of repeating patterns

2. Meter -­
the inferred metrical grid

3. The Time Sense

Source: Snyder, J. S., and E. W. Large. "Gamma-band Activity Reflects
the Metric Structure of Rhythmic Tone Sequences." Cog Brain Res 24 (2005):
117-126. Courtesy Elsevier, Inc., http://www.sciencedirect.com.
Used with permission.

Image of Salvador Dali's painting "The Persistence of Memory"

removed due to copyright restrictions.
See http://en.wikipedia.org/wiki/File:The_Persistence_of_Memory.jpg

Thursday, May 14, 2009

Rhythm: patterns of events in time

What is rhythm? Perceived patterns of events in time
What constitutes an event? What makes events salient
(accented)?
How many individual events can we distinguish (< 12/sec)?
Auditory sense and the time sense (supramodal)
– Perception of duration, weber fractions for time
Rhythmic pattern induction & expectation
– Rhythmic pattern invariance w. respect to tempo
Meter (regular underlying grid of accented/nonaccented
events)
Rhythmic hierarchies, rhythmic complexity
Small integer-ratios again: models (clock, oscillator, timing net)
Polyrhythms; analogy to polyphony
Interactions between melody & rhythm: accents
Rhythms: musical, body, and brain; kinesis
Thursday, May 14, 2009

Music notation: time durations

Fig. 2.2 in Sethares, W. A. Rhythm and Transforms.

Springer, 2007. ISBN: 9781846286391. Preview in Google Books.

Thursday, May 14, 2009

Tempo (absolute timescale, in beats/minute)

Slow Moderate Fast

<40 - 76 80 - 116 120 - 206+

Adagio ( = 60) Moderato ( = 90) Allegro ( = 120)

Beats Per Minute

Figure by MIT OpenCourseWare.

Thursday, May 14, 2009

 Tempo
Ranges of events; intervals from 50 ms to 2 sec
Too short: events fuse
Too long: successive events don't cohere, interact
Pitch (> 30 Hz); infra-pitch (10-30 Hz); rhythm (<
10 Hz)
For a brisk tempo of 120 bpm, 2 Hz,
 a quarter note is 500 msec (2 Hz)
an eighth note is 250 msec (4 Hz)
a sixteenth note is 125 ms (8 Hz)
a 32nd note is 62 ms (16 Hz)

Thursday, May 14, 2009

Rhythm: general observations I

• Different levels of temporal organization
– Handel’s basketball game analogy:
• Patterning
– Rhythm: perception of grouping & ordering of events
• Perceptual groupings of events in time create perceived
rhythmic patterns
• Temporal pattern expectancies create groupings
– pattern repetition and
– similar patterns of salient auditory contrasts (accents)
• Underlying temporal framework
– (metrical grid, meter, tempo)

Thursday, May 14, 2009

 Rhythm: recurring patterns of events in time

Every repeating pattern creates an
expectancy of its continuation

Figure by MIT OpenCourseWare.

Thursday, May 14, 2009

Every repeating pattern creates an

expectancy of its continuation

Further, there is a “chunking” of the repeating

pattern (the invariant pattern becomes an
object)

Figure by MIT OpenCourseWare.

Thursday, May 14, 2009

Rhythm generation demonstration

• Repeating patterns of events

• Drum score representation
• Synthesizer

10

Thursday, May 14, 2009

Acoustical grouping � � � (Snyder, Music & Memory)

Source: Snyder, Bob. Music and Memory.

Cambridge, MA: MIT Press, 2000. Courtesy of MIT Press.
Used with permission.
Thursday, May 14, 2009

Melodic & rhythmic grouping � (Snyder, Music & Memory)

Source: Snyder, Bob. Music and Memory.

Cambridge, MA: MIT Press, 2000. Courtesy of MIT Press.
Used with permission.

Thursday, May 14, 2009

Temporal grouping � � � (Snyder, Music & Memory)

Source: Snyder, Bob. Music and Memory.

Cambridge, MA: MIT Press, 2000.
Courtesy of MIT Press. Used with permission.

Thursday, May 14, 2009

Repetition
of a
rhythmic
pattern
establishes
the pattern Image removed due to copyright restrictions.
a) Two measure rhythmic pattern.
b) Complete 2-bar pattern, followed by a repetition of the complete pattern.
c) Complete 2-bar pattern, followed by two repetitions of the 2nd measure.
d) Complete 2-bar pattern, followed by two repetitions of the 2nd measure in reverse.
e) Complete 2-bar pattern; unique 3rd measure, and then a repetition of the 2nd measure.

Music Theory,

Thursday, May 14, 2009

Necklace notation: cyclical repeating patterns

Fig. 2.6 in Sethares, W. A. Rhythm and Transforms.

Springer, 2007. ISBN: 9781846286391.
Preview in Google Books.

Thursday, May 14, 2009

Necklace notation: cyclical repeating patterns

Fig. 2.4 in Sethares, W. A. Rhythm and Transforms. Springer, 2007. ISBN: 9781846286391. Preview in Google Books.

see Sethares, 2007. Necklace notation: Safi al-Din al-Urmawi 13th c. Bagdad Book of Cycles
Thursday, May 14, 2009

Necklace notation: cyclical repeating patterns

Fig. 2.5 in Sethares, W. A. Rhythm and Transforms. Springer, 2007. ISBN: 9781846286391.
Preview in Google Books .

Sethares, 2007
Thursday, May 14, 2009

Memory processes generate musical context

Tonality induction -- repetition of particular notes & sets of harmonics that
establishes a tonal expectation through which all subsequent
incoming tonal patterns are processed -- establishment of the tonic

Rhythmic induction -- repetition of patterns of accented and unaccented

events that establishes a temporal pattern of expectation for
subsequent events

Both kinds of induction operate on similarities and contrasts between

previous and subsequent sounds & events
OLD + NEW heuristic:
1) OLD incoming patterns similar to previous ones build up
the images of previous ones, confirm +
strengthen expectations, create relaxation
2) NEW different patterns create contrasts that violate
expectations established from previous inputs,
create tension
3) degree of contrast (distance in perceptual space) determines
the degree of tension created/resolved
Thursday, May 14, 2009

Hierarchy & time order (Snyder, Music & Memory, MIT Press, 2000)

Source: Snyder, Bob. Music and Memory.

Cambridge, MA: MIT Press, 2000.
Courtesy of MIT Press. Used with permission.

Thursday, May 14, 2009

Detection of arbitrary periodic patterns

Periodic patterns invariably build up in
delay loops whose recurrence times equals
the period of the pattern and its multiples.

0
1
2
3

1 = 11 ms = recurrence time
of input pattern10101100101
Input pattern

1010110010110101100101101011001011010...
Thursday, May 14, 2009

 Temporal coding of rhythm

S"mulus‐driven temporal pa3erns of spikes encode event structures
• Exist at the cor-cal level for periodici-es < 15 Hz
• Can directly encode rhythmic pa<erns
• Amenable to processing via recurrent -ming nets (RTNs)
• Chunk recurrent pa<erns of events to create rhythmic expectancies
∆ quality 

All time delays present no ∆

τ0 Time patterns reverberate
through delay loops
τ1
Recurrent,
τ2
indirect inputs ∆ pitch
τ3

Coincidence ∆ -mbre
units

Direct inputs
Figure by MIT OpenCourseWare.
Input time sequence
∆ dura-on

Thursday, May 14, 2009

 In addition to rhythmic patterning,
we seem to infer an underlying metrical
grid to the stream of events
(e.g. inferences that allow us to tap our
fingers or toes to a beat or to keep time
with the music)

This perception of an underlying metrical

order is important for coordination of
musicians playing in groups.

Meter serves as a temporal context that is

somewhat independent of individual events
(somewhat like the tonic vis-a-vis melody)
Thursday, May 14, 2009

 2:

3:

4:

6:

Meter and Accent

The recurrent groups of pulsations are called meters: for example, duple meter, triple
meter, and quadruple meter. The beats within the measures are counted and accented:

2: one, two | one, two |

3: one, two, three | one, two, three |
4: one, two, three, four | one, two, three, four |
6: one, two, three, four, five, six |

Figure by MIT OpenCourseWare.

Thursday, May 14, 2009

Meter (e.g. 4 pulses per measure, accent)

Definition: The number of pulses between the more or

less regularly recurring accents
 (Cooper and Meyer , 1960).
Most authors define meter similarly, as somehow
dependent upon (and perhaps contributing to)
patterns of accent.
Zuckerkandl (1956), however, views meter as a series
of "waves" that carry the listener continuously from
one beat to the next. For him, they result not from
accentual patterns but simply and naturally from the
constant demarcation of equal time intervals.
http://www.music.indiana.edu/som/courses/rhythm/illustrations
Thursday, May 14, 2009

Pulse & the metrical grid (meter)

Source: Snyder, Bob. Music and Memory.

Cambridge, MA: MIT Press, 2000. Courtesy of MIT Press.
Used with permission.

Thursday, May 14, 2009

Pulse

• Definition: A series of regularly recurring, precisely

equivalent stimuli ( Cooper and Meyer , 1960).

• According to Parncutt (1987), a chain of events,

roughly equally spaced in time.

http://www.music.indiana.edu/som/courses/rhythm/illustrations
Thursday, May 14, 2009

Visual grouping
Dember & Bagwell, 1985, A history of perception, Topics in the History of Psychology, Kimble & Schlesinger, eds.

Figure by MIT OpenCourseWare.

Thursday, May 14, 2009

Accent causes
grouping which
determines
perceived rhythmic
pattern

Rhythm is a
perceptual
attribute

Source: Handel, S. Listening: An Introduction to

the Perception of Auditory Events. Cambridge,
MA: MIT Press, 1989.
Courtesy of MIT Press. Used with permission.
Thursday, May 14, 2009

Factors that cause events to be accented:

�auditory contrast, salience
• note duration
• note intensity
• sharpness of attack
• duration of silence preceding it
• contrast: melodic contour/ pitch change
• regularity of timing (accented beats are "on time")
• position within a metrical organization
• According to Cooper & Meyer (1960), an accented tone
must be set off from the rest of the series in some way (i.e.
a salient contrast)
Thursday, May 14, 2009

Expressive timing & expectation

expressive timing Definition:

Music psychologists' term for the

deviations from a strictly uniform
pulse that occur in live performance.
These deviations most commonly
occur near the ends of phrases and
other grouping units. See Todd
(1985).

http://www.music.indiana.edu/som/courses/rhythm/illustrations
Thursday, May 14, 2009

 Meter and beat induction

Source: Palmer, C., and C. L. Krumhansl.

"Mental representations for musical meter."
J Exp Psychol Hum Percept Perform 16, no. 4 (Nov 1990): 728-741.
• From Thinking in Sound Courtesy of the American Psychological Association.

Thursday, May 14, 2009

rhythmic, metrical dissonance

• metrical dissonance Definition: According to

Krebs (1987), a situation in which the pulses in two
metrical levels are not well aligned, either because
the duration of the pulses in one level is not an
integral multiple or division of the duration of the
pulses in the other level, or because the pulses in
one level are displaced by some constant interval
from those in the other level. See also Yeston's
rhythmic dissonance .

http://www.music.indiana.edu/som/courses/rhythm/illustrations
Thursday, May 14, 2009

Event-related potentials & violations of temporal expectation

(notes, chords, beats, words (phonetic, semantic), many other levels of expectation)

Photo and graph of EEG/ERP removed due to copyright restrictions.

See: http://www.musicianbrain.com/methods.php#methods

Thursday, May 14, 2009

Thursday, May 14, 2009

Snyder & Large experiments on beat induction

J.S. Snyder, E.W. Large / Cognitive Brain Research 24 (2005) 117–126 119

Fig. 1. Pure-tone (262 Hz, 50 ms duration) stimulus patterns are shown with inter-onset intervals of 390 ms (above) and schematized metrical accent
representations (below). The periodic control condition consisted of isochronous tones designed to elicit a simple pulse perception (A). The binary control
condition consisted of alternating loud and soft tones, designed to elicit a duple meter perception (B). The omit-loud condition consisted of the binary control
pattern with missing loud tones on 30% of two-tone cycles (C). The omit-soft condition consisted of the binary control pattern with missing soft tones on 30%
of two-tone cycles (D).
35
Source: Snyder, J. S., and E. W. Large. "Gamma-band Activity Reflects the Metric
Structure of Rhythmic Tone Sequences." Cog Brain Res 24 (2005): 117-126.
Thursday, May 14, 2009 Courtesy Elsevier, Inc., http://www.sciencedirect.com. Used with permission.

 120 J.S. Snyder, E.W. Large / Cognitive Brain Research 24 (2005) 117–126

Source: Snyder, J. S., and E. W. Large.

"Gamma-band Activity Reflects the Metric
Structure of Rhythmic Tone Sequences."
Cog Brain Res 24 (2005): 117-126.
Courtesy Elsevier, Inc., http://www.sciencedirect.com.
Used with permission.

Figure 2. Process to calculate evoked and gamma-band activity (GBA).

Thursday, May 14, 2009

 Figure 4. Courtesy of University of Finance and Management, Warsaw. Used with permission.
(a) Time-frequency representation of the evoked and induced GBA results, averaged over all subjects. Tone onset occurs at
zero and 390 ms. (b) Comparison of induced/evoked peak activity in the presence and absence of loud and soft tones.

Thursday, May 14, 2009

 J.S. Snyder, E.W. Large / Cognitive Brain Research 24 (2005) 117–126

124

Source: Snyder, J. S., and E. W. Large. "Gamma-band Activity Reflects the Metric Structure of
Rhythmic Tone Sequences." Cog Brain Res 24 (2005): 117-126.
Courtesy Elsevier, Inc., http://www.sciencedirect.com. Used with permission.
Figure 7. Tone omissions: induced and evoked GBA.
Thursday, May 14, 2009

 Figure 5.
Perturbed stimuli; ‘x’ represents tone onset.
Courtesy of University of Finance and Management,
Warsaw. Used with permission.
Thursday, May 14, 2009

 Evoked GBA

Figure 6. Courtesy of University of Finance and Management, Warsaw. Used with permission.
Time-frequency representation of the evoked and induced GBA in response to early, late, or on-time tones averaged over all
subjects. The white dashed line represents where a tone was expected. (a) Evoked activity is predicted by the presence of
tones. The white box highlights an exception, activity where the tone was expected in the case of an early tone. (b) The white
box indicates a peak in the induced activity where the tone was expected for the case of late tones.

Thursday, May 14, 2009

 Induced GBA

Figure 6. Courtesy of University of Finance and Management, Warsaw. Used with permission.
Time-frequency representation of the evoked and induced GBA in response to early, late, or on-time tones averaged over all
subjects. The white dashed line represents where a tone was expected. (a) Evoked activity is predicted by the presence of
tones. The white box highlights an exception, activity where the tone was expected in the case of an early tone. (b) The white
box indicates a peak in the induced activity where the tone was expected for the case of late tones.

Thursday, May 14, 2009

 SUMMARY
Evoked GBA appears to represent sensory processing
as predicted by the presence of tones, much like the
MLR. Induced GBA may reflect temporally precise ex­
pectancies for strongly and weakly accented events in
sound patterns. Moreover, induced GBA behaves in a
manner consistent with perception-action coordination
studies using perturbed temporal sequences. Taken
together, the characteristics of induced GBA provide
evidence for an active, dynamic system capable of
making predictions (i.e., anticipation), encoding metri­
cal patterns and recovering from unexpected stimuli.
GBA appears to be a useful neuroelectric correlate
of rhythmic expectation and may therefore reflect pulse
perception. Due to the anticipatory nature of GBA, it
may be supposed there is an attentional dependence.
Future research should aim to manipulate attentional
state, localize neural sources and further probe the
role of induced GBA in meter perception.

Thursday, May 14, 2009 Courtesy of University of Finance and Management, Warsaw. Used with permission.

Syncopation - violation of metrical expectations

Image removed due to copyright restrictions.

Definition of syncopation with some musical examples.
From Jones, G. T. Music Theory. New York, NY: Barnes and Noble Books, 1974.

Music Theory, Thad. Jones

Thursday, May 14, 2009

Rhythmic streaming (segregation/fusion of rhythmic

• African xylophone music

• Timbre effects

• Pitch difference

• Competition of frequency separations

47

Thursday, May 14, 2009

Rhythmic elaboration -subdividing time intervals

Figure by MIT OpenCourseWare.

Smulevitch & Povel (2000) in Rhythm: Perception & Production, Desain & Windsor eds

Thursday, May 14, 2009

Rhythmic

Hierarchy

Source: Handel, S. Listening: An Introduction to

the Perception of Auditory Events. Cambridge,
MA: MIT Press, 1989.
Courtesy of MIT Press. Used with permission.

Handel
Thursday, May 14, 2009

Rhythmic Hierarchy

Source: Handel, S. Listening: An Introduction to

the Perception of Auditory Events. Cambridge,
MA: MIT Press, 1989.
Courtesy of MIT Press. Used with permission.
Handel
Thursday, May 14, 2009

 Source: Handel, S. Listening: An Introduction to

the Perception of Auditory Events. Cambridge,
MA: MIT Press, 1989.
Courtesy of MIT Press. Used with permission.

Handel
Thursday, May 14, 2009

Polyrhythms

Source: Handel, S. Listening: An Introduction to

the Perception of Auditory Events. Cambridge,
MA: MIT Press, 1989.
Courtesy of MIT Press. Used with permission.
Handel
Thursday, May 14, 2009

 Source: Handel, S. Listening: An Introduction to

the Perception of Auditory Events. Cambridge,
MA: MIT Press, 1989.
Courtesy of MIT Press. Used with permission.

Source: Handel, S. Listening: An Introduction to

the Perception of Auditory Events. Cambridge,

MA: MIT Press, 1989.

Courtesy of MIT Press. Used with permission.

Thursday, May 14, 2009

 Polyrhythms (polyrhythms:rhythm::polyphony:melody)
Source: Handel, S. Listening: An Introduction to
the Perception of Auditory Events. Cambridge,
MA: MIT Press, 1989.
Courtesy of MIT Press. Used with permission.

Conlon Nancarrow

Handel
Thursday, May 14, 2009
Rhythm & Grouping

• Three examples from

• Bregman & Ahad
• Auditory Scene Analysis CD
• African xylophone music
�interference between rhythmic patterns
� separation of patterns via pitch differences
� separation of patterns via timbral diffs
�
Conflicting rhythms interfere unless the events
can be separated out in separate streams

Thursday, May 14, 2009

Metrical vs. rhythmic phrases (rel. independence)

(Snyder, Music & Memory)

Source: Snyder, Bob. Music and Memory.

Cambridge, MA: MIT Press, 2000. Courtesy of MIT Press.
Used with permission.

Thursday, May 14, 2009

Major points -- rhythm

Rhythm involves perception of temporal patterns of events

Recurring patterns group into chunks that create expectations of

future temporal occurences of events (rhythmic pattern
induction)

Rhythmic grouping occurs on the same timescale as

melodic grouping.

We also infer a metrical grid that involves a regular set of

timepoints (pulse, tatum) and a regular pattern of accented/
unaccented events (meter). (Metrical induction)

Expectations generated from rhythmic grouping and metrical

induction processes can be manipulated for tension-relaxation
effect.
Thursday, May 14, 2009

 Time, memory, and anticipation

Image of Salvador Dali's painting "The Persistence of Memory"

removed due to copyright restrictions.
See http://en.wikipedia.org/wiki/File:The_Persistence_of_Memory.jpg

Thursday, May 14, 2009

Temporal integration windows � (Snyder, Music & Memory)

Source: Snyder, Bob. Music and Memory.

Cambridge, MA: MIT Press, 2000.
Courtesy of MIT Press. Used with permission.

Thursday, May 14, 2009

Timescales
& memory�

(Snyder, Music & Memory)

Source: Snyder, Bob. Music and Memory.

Cambridge, MA: MIT Press, 2000.
Courtesy of MIT Press. Used with permission.

Thursday, May 14, 2009

Memory
& grouping �
(Snyder, Music & Memory)

Source: Synder, B. Music and Memory.

Cambridge, MA: MIT Press,
2000.Courtesy of MIT Press.
Used with permission.

Thursday, May 14, 2009

Memory processes generate musical context

Tonality induction -- repetition of particular notes & sets of harmonics that
establishes a tonal expectation through which all subsequent
incoming tonal patterns are processed -- establishment of the tonic

Rhythmic induction -- repetition of patterns of accented and unaccented

events that establishes a temporal pattern of expectation for
subsequent events

Both kinds of induction operate on similarities and contrasts between

previous and subsequent sounds & events
OLD + NEW heuristic:
1) OLD incoming patterns similar to previous ones build up
the images of previous ones, confirm +
strengthen expectations, create relaxation
2) NEW different patterns create contrasts that violate
expectations established from previous inputs,
create tension
3) degree of contrast (distance in perceptual space) determines
Thursday, May 14, 2009

 Time, memory, and anticipation

Thursday, May 14, 2009

Time
What is time? Newtonian & Bergsonian time
The perception of time
Duration, succession, and perspective
Relativity of time
Constant Weber fraction for time estimation
Aging & time perception (internal clocks slow down)
Duration and event-density
Learning & temporal prediction (anticipation)
Brains as temporal prediction machines
Models of time (interval) perception & production
Clock models -- accumulators (hourglass)
Oscillator models (pendulum)
Delay-detectors and static representations of time
Rhythmic hierarchies, simple ratios, and groupings
Temporal memory traces (delay loops, cyclochronism)

Thursday, May 14, 2009

Time
"time…does not exist without changes." Aristotle, Physics, IV

Time as an absolute world-coordinate (Newtonian time) vs. time as epistemic

change (psychological, Bergsonian time)

"A man in sound sleep, or strongly occupy'd with one thought, is insensible of time… Whenever we
have no successive peceptions, we have no notion of time, even tho' there be a real succession
in the objects…time cannot make its appearance to the mind, either alone, or attended with a
steady unchangeable object, but is always discovered by some perceivable succession of
changeable objects." Hume as quoted in Fraisse, pp. 3-4

Measurement of time
How is time measured, psychologically, by the neural mechanisms and informational
organizations that constitute our minds?

Thursday, May 14, 2009

Duration 60 BPM

Our sense of the length of time (Fraisse, 1962, The Psychology of Time)
152 BPM
Constant Weber fractions for interval estimation
Errors are proportional to the interval estimated
Weber's law for timing; jnd's on the order of 8-12%
depending on modality (hearing, touch, vision)
Temporal prediction of reward in conditioning
(Scalar timing intimately related to the response latency in conditioning when interval
between stimulus and reward are varied, see R. Church, A Concise Introduction to Scalar
Timing Theory, 2003. See also Fraisse's (1963) discussion of Pavlov and Popov
cyclochronism model)
Some general observations (Fraisse via Snyder, Music & Memory):
Filled time durations appear shorter than empty ones
Rate of novel events makes durations appear shorter
(monotonous durations are experienced as longer, but remembered as shorter)
Aging: young children overestimate durations; older adults underestimate durations
(A systematic change in internal timing mechanisms with age? cf �absolute pitch)
Implications for music: pieces with high event densities go faster; those
with low ones seem to take forever; duration is in the mind of the
beholder and his/her expectations
Thursday, May 14, 2009

Beat
induction
and
duration
discrimination

Weber's
Law Image removed due to copyright restrictions.

Graph illustrating Weber's Law. See Fig. 4.13 in Jones and Yee,

"Attending to auditory events: the role of temporal organization."

In Thinking in Sound. Edited by E. Bigand and S. McAdams.

New York, NY: Oxford University Press, 1993. ISBN: 9780198522577.

Thursday, May 14, 2009

Succession
Time order:
before and after (Fraisse, Snyder)

Our recollection of time order

depends on memory mechanisms,
how distant in the past were the
events

Representation of order in long-term

memory is poor
LTM is massively parallel, not serial
Time order within chunks is better
preserved than between them
Primacy and recency: first and last
elements in a chunk best
remembered, most salient

Thursday, May 14, 2009

Perspective: Past, present, future

Mediated by different psych/brain mechanisms
Past: long term memory
Present: working memory
Future: anticipation, planning

Music (like sports) focuses our minds on the

present, on events that have occurred in the last
few seconds to minutes.

Thursday, May 14, 2009

Mechanisms of timing and temporal processing

• Temporal contiguity models of learning

• Clock models
– Switched accumulator, e.g. hourglass
– Explicit measurement of time durations
– Ordering of durations by magnitude
• Time delay detectors/generators
– Array of tuned delay elements, detectors, oscillators
– Explicit measurement of time durations; storage of patterns
– Generators of time delays (timers)
• Rhythmic hierarchies (Jones)
– well-formed patterns create strong expectations
• Temporal memory trace
– Timeline of events stored in reverberating memory
– Readout of events & (timing of) their consequences

Thursday, May 14, 2009

Temporal expectations on different timescales

• Pitch: repetitions on microtemporal

� �timescales (200 usec to 30 ms)
• Infra-pitch: not well defined, repetitions
� �with periods 30-100 ms
• Rhythms: patterns of individuated events
�with periods 100 ms to several secs
• Longer temporal expectations (> few secs)

Thursday, May 14, 2009

 Metrical and nonmetrical patterns (cf. tonal & atonal melodies)

Image removed due to copyright restrictions.

See Fig. 4.8 in Jones and Yee, "Attending to auditory events:
the role of temporal organization." In Thinking in Sound.
Edited by E. Bigand and S. McAdams. New York, NY:
Oxford University Press, 1993. ISBN: 9780198522577.

Jones & Yee,

Attending to auditory events:
the role of temporal organization
in Thinking in Sound

Thursday, May 14, 2009

 Temporal reproductions are better for well-formed temporal patterns

Image removed due to copyright restrictions.

See Fig. 4.9 in Jones and Yee, "Attending to auditory events:

the role of temporal organization." In Thinking in Sound.

Edited by E. Bigand and S. McAdams. New York, NY:

Oxford University Press, 1993. ISBN: 9780198522577.

Thursday, May 14, 2009

Higher-order (longer-range) metrical patterns

Image removed due to copyright restrictions.

See Fig. 4.7 in Jones and Yee, "Attending to auditory events:

the role of temporal organization." In Thinking in Sound.

Edited by E. Bigand and S. McAdams. New York, NY:

Oxford University Press, 1993. ISBN: 9780198522577.

Thursday, May 14, 2009

 Hierarchical & nonhierarchical ratios of event timings

Image removed due to copyright restrictions.

See Fig. 4.10 in Jones and Yee, "Attending to auditory events:

the role of temporal organization." In Thinking in Sound.

Edited by E. Bigand and S. McAdams. New York, NY:

Oxford University Press, 1993. ISBN: 9780198522577.

Thursday, May 14, 2009

Clock & hierarchical models of beat perception

Image removed due to copyright restrictions.

See Fig. 4.11 in Jones and Yee, "Attending to auditory events:

the role of temporal organization." In Thinking in Sound.

Edited by E. Bigand and S. McAdams. New York, NY:

Oxford University Press, 1993. ISBN: 9780198522577.

Thursday, May 14, 2009

Mechanisms of timing and temporal processing

• Temporal contiguity models of learning

• Clock models
– Switched accumulator, e.g. hourglass
– Explicit measurement of time durations
– Ordering of durations by magnitude
• Time delay detectors/generators
– Array of tuned delay elements, detectors, oscillators
– Explicit measurement of time durations; storage of patterns
– Generators of time delays (timers)
• Rhythmic hierarchies (Jones)
– well-formed patterns create strong expectations
• Temporal memory trace
– Timeline of events stored in reverberating memory
– Cylcochronism (Popov, see Fraisse, memory store is itself temporal)
– Readout of events & (timing of) their consequences
Thursday, May 14, 2009

Warren:
Holistic &
analytic
sequence
recognition

holistic:
temporal
compounds
Image removed due to copyright restrictions.

(cohere into See Fig. 3.2 in Warren, "Perception of acoustic sequences:

unified global integration vs. temporal resolution." In Thinking in Sound.

Edited by E. Bigand and S. McAdams. New York, NY: Oxford University

patterns) Press, 1993. ISBN: 9780198522577.

analytic:
explicit ID
of elements
and orders

Thursday, May 14, 2009

Timescale similarities & differences of temporal processing

• On all timescales:
– mechanisms for internalizing timecourses of

events, for building up temporal patterns

• Differences between timescales

– Pitch: support of multiple patterns (pitch

mechanism low harmonics) => temporal

"transparency", non-interference

– Rhythm: interference between patterns unless

separated into different streams
– another way of thinking about this is that for
rhythm stream formation mechanism is not based
on periodicity alone

Thursday, May 14, 2009

 Licklider�s (1951) duplex model of pitch perception

Time-delay nets

Licklider�s binaural triplex model

Figure by MIT OpenCourseWare.

Image removed due to copyright restrictions.

Frequency
x
Pe
rio
di
cit
y

J.C.R. Licklider (1959) “Three

AuditoryTheories” in Psychology: A Study
of a Science, Vol. 1, S. Koch, ed., McGraw-
Hill, pp. 41-144.
Cochlea
Figure by MIT OpenCourseWare.

Thursday, May 14, 2009

 Neural timing nets

FEED-FORWARD TIMING NETS RECURRENT TIMING NETS

• Temporal sieves • Build up pattern invariances
• Extract (embedded) similarities • Detect periodic patterns
• Multiply autocorrelations • Separate auditory objects

Si(t) two sets

All time delays present
of input
Sj(t) spike trains 0 Time patterns reverberate
hrough delay loops
Si(t) Sj(t - ) 1 t
Recurrent,
individual
indirect inputs
multiplicative 2
term
3
� Si(tm) Sj(tm - t)

convolution
time-series Coincidence
term m units

Time Direct inputs

t
Input time sequence

Relative delay 

Thursday, May 14, 2009

 Is a time-domain strategy possible?

Effect of different F0s in the time domain

Vowel [ae]

F0 = 100 Hz

Vowel [er]

F0 = 125 Hz

Double vowel

[ae]+[er]

0 10 20 30 40 50
Time (msec)
Thursday, May 14, 2009

Auditory "pop-out"
phenomena suggest
Last 2 periods - first 2
a period-by-period

transient
disparity

ongoing
disparity

Thursday, May 14, 2009

 Figure by MIT OpenCourseWare.

Thursday, May 14, 2009

Detection of arbitrary periodic patterns

Periodic patterns invariably build up in
delay loops whose recurrence times equals
the period of the pattern and its multiples.

0
1
2
3

1 = 11 ms = recurrence time
of input pattern10101100101
Input pattern

1010110010110101100101101011001011010...
Thursday, May 14, 2009

 All time delay present

Time patterns reverberate through

delays loops

Recurrent, indirect inputs

Coincidence units

Direct inputs

Input waveform

Figure by MIT OpenCourseWare.

Thursday, May 14, 2009

 Recurrent
timing net
with single

Revised buildup rule:

Min(direct, circulating)
plus a fraction of their
absolute difference

Source: Cariani, P. "Neural Timing Nets."

Neural Networks 14 (2001): 737-753.
Courtesy Elsevier, Inc.,
http://www.sciencedirect.com.
Used with permission.

Thursday, May 14, 2009

 Error-adjustment rule:

H(t) = H(t-tau) + Btau[X(t)-H(t-tau)]

Loop-dependent scaling of adj rate:

Btau = tau/33 ms

Thursday, May 14, 2009

Thursday, May 14, 2009

Thursday, May 14, 2009

Thursday, May 14, 2009

Thursday, May 14, 2009

Tonal & rhythmic contexts

Tonality induction:
establishment of a tonic
establishment of tonal system: key, mode, set of pitches
establishment of harmonic relations

Western tonal music:

Relations of notes to the tonic
Relations of notes to the triad that defines the key (I)
harmonic center
Relations of chords to I triad & tonic -- chord progressions
Distance in perceptual similarity
Tension-resolution + movement between the two
Relations of different keys and key modulations
Movements between keys, tension-resolution, larger
structures & rhythms of harmonic movement

Thursday, May 14, 2009

Build-up and separation of two auditory objects

Two vowels with different fundamental frequencies (F0s) are added together
and passed through the simple recurrent timing net. The two patterns build up
In the delay loops that have recurrence times that correspond to their periods.

Characteristic delay channel (ms)

0 10 20 30 40 50
Time (msec)

Vowel [ae]
F0 = 100 Hz
Period = 10 ms

Vowel [er]
F0 = 125 Hz
Period = 8 ms

Time (ms)
Thursday, May 14, 2009
 Fragment from G. Ligeti's Musica Ricercata

-1

1 Signal rms envelope

0.5

0
0 1 2 3 4 5 6 7 8 9 sec

This image is from the article Cariani, P. "Temporal Codes, Timing Nets, and Music Perception."
Journal of New Music Research 30, no. 2 (2001): 107-135. DOI: 10.1076/jnmr.30.2.107.7115.
This journal is available online at http://www.ingentaconnect.com/content/routledg/jnmr/

Thursday, May 14, 2009

 Fragment from G. Ligeti's Musica Ricercata

1

-1

Signal rms envelope

1

0.5

0
0 1 2 3 4 5 6 7 8 9 sec

Autocorrelogram
200

150

100

50

100 200 300 400 500 600 700 800

TIme re: signal onset (samples)

Recurrent timing net

200

150

100

50

100 200 300 400 500 600 700 800

TIme re: signal onset (samples)

This image is from the article Cariani, P. "Temporal Codes, Timing Nets, and Music Perception."
Journal of New Music Research 30, no. 2 (2001): 107-135. DOI: 10.1076/jnmr.30.2.107.7115.
This journal is available online at http://www.ingentaconnect.com/content/routledg/jnmr/
Thursday, May 14, 2009

 Autocorrelogram
200

150
Delay (samples)

100

50

100 200 300 400 500 600 700 800

TIme re: signal onset (samples)

Profile of mean signal values in delay channels Profile of std. deviations in delay channels
44 40

42 35
40
30
38
25
36
20
34

32 15
0 50 100 150 200 0 50 100 150 200
Delay (samples) Delay (samples)

This image is from the article Cariani, P. "Temporal Codes, Timing Nets, and Music Perception."
Journal of New Music Research 30, no. 2 (2001): 107-135. DOI: 10.1076/jnmr.30.2.107.7115.
This journal is available online at http://www.ingentaconnect.com/content/routledg/jnmr/

Thursday, May 14, 2009

 Autocorrelogram Ligeti envelope (fragment end)

Profile of mean signal values in delay channels
44

42

40

38

36
6
Loop delay = 88
34
4
32
0 50 100 150 200
Delay (samples) 2

0
Recurrent timing net
6
Loop delay = 134
1.5
4

0
1
8 Loop delay = 178
Time (samples)
6

2
0.5
0
0 50 100 150 200 700 750 800 850
Delay channel (samples) Time (samples)

Response of a recurrent timing network to the Ligeti fragment.

H( j,i +j) = max (X( 1,i ), X( 1,i )*H( j,i )*(1+j/100)) where X is the in
envelope of the Ligeti fragment, and H is the value of the signal in delay loop j (firs
index) at time t (second index). The buildup factor (1+j/100) depends on the duration
of the delay loop (i.e. equal to j samples). The mean signal value H in the delay
channels over the last 200 samples (thicker line) and over the whole fragment (thin
line) are shown in the top right line plot. The waveforms that are built up in the three
most activated delay loops are shown above. The results, not surprisingly resemble
those obtained with the running autocorrelation. The sampling rate of the
signal was approximately 10 Hz.

This image is from the article Cariani, P. "Temporal Codes, Timing Nets, and Music Perception."
Journal of New Music Research 30, no. 2 (2001): 107-135. DOI: 10.1076/jnmr.30.2.107.7115.
This journal is available online at http://www.ingentaconnect.com/content/routledg/jnmr/
Thursday, May 14, 2009

MIT OpenCourseWare
http://ocw.mit.edu

HST.725 Music Perception and Cognition

Spring 2009

For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.