Perception of Emotion in Sounded and Imagined Music

Abstract

WE STUDIED THE EMOTIONAL RESPONSES BY MUSICIANS to familiar classical music excerpts both when the music was sounded, and when it was imagined.We used continuous response methodology to record response profiles for the dimensions of valence and arousal simultaneously and then on the single dimension of emotionality. The response profiles were compared using cross-correlation analysis, and an analysis of responses to musical feature turning points, which isolate instances of change in musical features thought to influence valence and arousal responses. We found strong similarity between the use of an emotionality arousal scale across the stimuli, regardless of condition (imagined or sounded). A majority of participants were able to create emotional response profiles while imagining the music, which were similar in timing to the response profiles created while listening to the sounded music.We conclude that similar mechanisms may be involved in the processing of emotion in music when the music is sounded and when imagined.

Full text

Music Perception VOLUME 27, ISSUE 5, PP. 399–412, ISSN 0730-7829, ELECTRONIC ISSN 1533-8312 © 2010 BY THE REGENTS OF THE UNIVERSITY OF CALIFORNIA. ALL
RIGHTS RESERVED
. PLEASE DIRECT ALL REQUESTS FOR PERMISSION TO PHOTOCOPY OR REPRODUCE ARTICLE CONTENT THROUGH THE UNIVERSITY OF CALIFORNIA PRESS’S
RIGHTS AND PERMISSIONS WEBSITE
, HTTP://WWW.UCPRESSJOURNALS.COM/REPRINTINFO.ASP. DOI:10.1525/MP.2010.27.5.399
Subdivision of Long Intervals 399
B
RIAN J. LUCAS
Bucknell University
E
MERY SCHUBERT
University of New South Wales, Sydney, Australia
A
NDREA R. HALPERN
Bucknell University
WE STUDIED THE EMOTIONAL RESPONSES BY MUSICIANS
to familiar classical music excerpts both when the music
was sounded, and when it was imagined. We used con-
tinuous response methodology to record response
profiles for the dimensions of valence and arousal simul-
taneously and then on the single dimension of emo-
tionality. The response profiles were compared using
cross-correlation analysis, and an analysis of responses
to musical feature turning points, which isolate instances
of change in musical features thought to influence
valence and arousal responses. We found strong simi-
larity between the use of an emotionality arousal scale
across the stimuli, regardless of condition (imagined or
sounded). A majority of participants were able to create
emotional response profiles while imagining the music,
which were similar in timing to the response profiles
created while listening to the sounded music. We conclude
that similar mechanisms may be involved in the pro-
cessing of emotion in music when the music is sounded
and when imagined.
Received November 11, 2008, accepted February 25, 2010.
Key words: musical imagery, emotion, continuous
response, cross-correlation, arousal
M
ENTAL IMAGERY HAS INTRIGUED COGNITIVE
psychologists for several reasons. On the one
hand, most people experience these quasi-per-
ceptual representations and report them as being very
vivid and powerful. For example, among Gabrielsson’s
(2010) analysis of reports of people’s strong experiences
in music, he found that some:
...may be elicited by ‘inner music’; that is, imagined
music ...Music is ‘heard’, it just comes for no obvious
reason and ‘sounds’ as clear and distinct as live music.
Some respondents realized that it was ‘only’ imagined
music, but for others the experience was so vivid that
they were surprised to learn that there was in fact no
sounding music present at all. (Gabrielsson, 2010, p. 559)
On the other hand, a mental process that in some ways
mimics a perceptual process is resource intensive and
thus not very efficient. However, despite this inefficiency,
there may be some kinds of information that are ideally
suited for this kind of representation. Kozhevnikov,
Kosslyn, and Shephard (2005) found that scientists, who
often have to think about how variables are related, were
particularly good at a task of spatial imagery. The same
tradeoff of efficiency for accuracy obtains for auditory
imagery, especially for music.
From a myriad of composers who insist that music is
composed in the mind before it is ever sounded, to the
average person who cannot get a melody line out of his
or her head, it is likely that most musicians and nonmu-
sicians alike have at one time or another experienced
such music images. A number of studies have shown that
these representations maintain features of sounded music,
such as pitch, tempo, melody, mode, and loudness. For
instance, Halpern (1988a) asked participants to identify
whether or not two lyrics were part of the same song and
observed that reaction times increased as the distance,
in beats, between the lyrics in the sounded tune increased.
These results suggested that participants temporally
scanned the musical image to locate the lyrics and this
demonstrates that at least one temporal aspect of music
is maintained in imagery. Halpern, Zatorre, Bouffard,
and Johnson (2004) showed that similarity ratings of
instrument sounds were nearly equivalent whether the
sounds were perceived or imagined (as were several areas
of neural activation), suggesting that another musical
feature, timbre, is represented in musical images.
Although it has been established that these musical
features can be represented in musical imagery, it is less
clear whether the emotion in the music also is represented.
We know very little about this, although we may get a
hint from self-report studies of persistent musical images
(“earworms”) that are often of preferred music (Bailes,
2007). To the extent that anyone voluntarily “replays” a
tune in his or her head, it is a reasonable assumption that
this is partly because the tune is conveying some affect
PERCEPTION OF EMOTION IN SOUNDED AND IMAGINED MUSIC
Music2705_05 5/13/10 2:31 PM Page 399

desired at the moment. And musicians who engage in
mental practice may perhaps be doing so partly to review
the emotional expression they wish to convey in actual
performance. However, with one exception noted below,
no one has examined this question under controlled
conditions.
Furthermore, we know that some structural features
communicate emotion. Hevner (1935) presented par-
ticipants with an adjective circle that clustered together
similar adjectives that represented a general emotion cat-
egory. Hevner found, for example, that the major mode
generally is associated with positive descriptors and the
minor mode with negative descriptors. Using a similar
checklist methodology, Motte-Haber (as cited in
Gabrielsson & Lindstrom, 2001) found that music with
faster tempi and higher event densities are rated as hap-
pier than music with slower tempi and lower event den-
sities. Other musical features that have been linked to
emotional descriptors include pitch and loudness. Higher
pitches have been associated with happiness and lower
pitches with sadness. Louder music and faster tempo have
been associated with excitement and softer music with
sleepiness (Gabrielsson & Lindstrom, 2001, Schubert,
2004). Other researchers have combined these basic fea-
tures. For instance, Gagnon and Peretz (2003) found that
both (major) mode and (faster) tempo elicited “happy”
judgments and vice versa for “sad,” with tempo being the
more salient influence.
If at least some of these features can be represented in
imagery, we propose that related emotional judgments
may also be represented. However, measuring emotional
judgment of imagined music presents challenges. Mental
imagery is a cognitively demanding process, as it requires
considerable working memory capacity. Making emo-
tional judgments during this retrieval might be difficult.
Thus, if an emotional judgment is elicited after the
retrieval, the respondent could be basing the judgment
on semantic associations to the music, or perhaps report-
ing socially desirable responses. It is difficult to verify the
accuracy of the memory representation of the piece, and
the retrospective judgment could be affected by that accu-
racy (or lack thereof).
It seems more desirable to assess emotional judgments
of music as the experience unfolds (we want to clarify
here that we are asking people to judge what the music
is communicating, not how the music makes one feel;
see Evans & Schubert, 2008). This approach allows a more
temporally fine-grained analysis of emotional response,
and also allows a mapping of emotional response to par-
ticular structural features in the music. Continuous
response (CR) methodology (Schubert, 2001) allows emo-
tional response to be measured in one or two dimensions
as the music unfolds. Valence and arousal have been rec-
ognized by several researchers (e.g., Nagel, Kopiez, Grewe,
& Altenmüller, 2007) as adequately capturing the gamut
of emotional response. Valence is an emotional dimen-
sion that ranges from negative emotions, such as sad and
angry, to positive emotions, such as happy and calm.
Arousal is a dimension that ranges from excited/ ener-
gized to sleepy/bored. Arousal can differentiate between
emotions with similar levels of valence. For example,
“sad” would be distinguished from “angry” by its low
level of arousal compared to anger’s high level of arousal
(Russell, 1980).
In a typical CR task, a listener is trained to indicate with
a computer mouse his or her response on the emotional
dimension of interest, as the music is played. The mouse
position in one- or two-dimensional space is recorded at
frequent, regular time intervals. The data can later be cor-
related with events in the musical stream. Although the
response is unlikely to be instantaneous (for example,
Schubert & Dunsmuir, 1999, found a lag of about 2–3 s
for emotional response to changes in loudness), these
responses are systematically related to musical features,
and can provide a fine-grained and real-time portrait of
emotional response to even complex music.
With this paradigm, we are in a position to ask whether
the type and time course of emotional judgments of
music are similar in sounded and imagined situations.
We can ask a participant first to listen to a piece of famil-
iar music and respond with valence and arousal ratings
(or emotionality ratings, as in our Experiment 2) as the
music unfolds. Then the participant repeats the task
while imagining the music. The response profile to the
sounded music and to the imagined music can then be
compared. Schubert, Evans, and Rink (2006) reported
a case study of one individual in this task. A professional
pianist listened to his own recording of a Chopin
Nocturne while making valence and arousal judgments.
He then repeated the procedure while imagining his
recording. The authors found that the continuous
response profile of the sounded music strongly resem-
bled that of the imagined music, although they also
found that the pianist gradually slowed his responses in
the imagined condition, perhaps due to the cognitive
load of retrieving the imagined music (Sweller, 1988,
2006). However, we do not know if this response was
idiosyncratic, confined to a professional musician, or
confined to imagining one’s own performance.
The current study built on this preliminary work by
asking a group of student musicians to make continu-
ous emotional judgments for excerpts of three different
orchestral pieces. We looked at the congruence of emo-
tional judgments with musical features that we predicted
400
Brian J. Lucas, Emery Schubert, & Andrea R. Halpern
Music2705_05 5/13/10 2:31 PM Page 400

a priori would engender specific responses, and also
compared the response profiles in the sounded and
imagined conditions. In Experiment 1, participants per-
formed the continuous response task using the dimen-
sions of valence and arousal. In Schubert’s review of the
emotion labels used for continuous response, he argued
that “arousal seems to be similar to emotional strength”
(2010, p. 241) and that more research was called for to
examine this similarity. In Experiment 2, participants
used the single dimension of emotionality, both to see
if a single global dimension could be extracted from the
music, and to allow us to compare this global attribute
to the more specific emotional labels of valence and
arousal. Experiment 2 also replicated the procedure of
Experiment 1 for several returning participants, in an
attempt to assess test-retest reliability of results after
several months.
We predicted that the continuous response time-series
profiles of the sounded and imagined conditions would
be structurally similar, which would demonstrate a
musician’s ability to extract emotion from a musical
image. The Schubert et al. study (2006) reported con-
sistent temporal drift, or lag, in the imagined responses
compared to the sounded responses, but with good
structural similarity between the sounded and imag-
ined time-series profiles of emotional response. After
correcting for any lagging that may occur, we predicted
that the imagined responses would be structurally con-
sistent with the sounded responses for experienced
musicians who were familiar with the music. An alter-
native prediction stems from the fact that the musicians
tested here, while experienced, were not as intimately
familiar with the pieces as was the pianist who had per-
formed the piece in the prior study. For these partici-
pants, performing this somewhat novel task of tracking
emotion might induce some additional cognitive load,
which could induce less correspondence between
sounded and imagined emotion profiles than Schubert
and colleagues had found.
Experiment 1
The aim of Experiment 1 was to investigate whether emo-
tional responses made continuously to music were sim-
ilar when the music was sounded compared to when it
was imagined. Participants first were given a brief tap-
ping task to establish their ability to keep time. In the
experimental task, continuous response methodology
was used to track participants’ emotional responses as
they judged three familiar pieces of classical music on
the dimensions of valence and arousal. This task included
a sounded condition and an imagined condition.
Method
PARTICIPANTS
Participants were 22 Bucknell University undergraduate
students (18–21 years old). Five individuals later were
excluded (for reasons explained in the Results section);
thus the study included data from 17 (13 female and 4
male) participants. The group consisted of students from
the Introductory Psychology participant pool (N = 18)
and music majors (N = 4) who were compensated with
course credit and movie tickets, respectively. To increase
the probability that participants would be adept in musi-
cal tasks and familiar with the classical music excerpts
used here, they were required to have a minimum of 8
years of private instrumental lessons. Experience ranged
from 8–17 yrs with an average of 10.81 years.
MATERIALS
The stimuli were excerpts of Romantic and late Classical
Western music of less than 1 min in length. Classical
music was chosen because of the large selection of word-
less pieces that avoid the complex interaction between
musical features and lyrics (Serafine, Crowder, & Repp,
1984). High levels of richness and musical feature vari-
ety within a short excerpt is a common feature of Romantic
and late Classical Western music not ordinarily found in
other styles of music likely to be familiar to the partici-
pants. In order to identify highly familiar pieces that con-
tain musical variability (changes in tempo, melody,
loudness), we developed a survey that included 16 clas-
sical pieces. The survey was administered to six Bucknell
University students with musical backgrounds. The three
pieces listed below were chosen for the experiment
because of high scores on scales measuring familiarity
and confidence in one’s ability to imagine the piece, and
musical feature variability.
1. Allegro con brio from Beethoven’s 5th Symphony in
C minor, Op. 67 (Classical Hits, Films for Humanities
and Sciences, performed by the Royal Philharmonic;
abbreviated B). Measures 1–63 (48 s) were expected to
occupy the low valence/high arousal and low valence/low
arousal quadrants of the two-dimensional emotion space
(see Figure 1). Its minor mode was predicted to evoke
various levels of negative valence and its extreme changes
in loudness were expected to produce varying levels of
arousal.
2. The Allegro from Mozart’s Serenade in G K525 (Eine
Kleine Nachtmusik) (Deutsche Grammophon/Polygram,
performed by the Berlin Philharmonic, M). Measures 1–28
(53 s) were expected to occupy the high valence/high
arousal and high valence/low arousal quadrants of the
two-dimensional emotion space. Its mode is major, which
was predicted to evoke positive valence, and its variations
Perception and Imagination of Emotion 401
Music2705_05 5/13/10 2:31 PM Page 401

in loudness and note density would produce various levels
of arousal.
3. Tchaikovsky’s Waltz of the Flowers from Nutcracker
Suite Op. 71 (Classical Hits, Films for Humanities and
Sciences, performed by the Royal Philharmonic, T.)
Measures 34–86 (51 s) were expected to occupy the high
valence/low arousal, and high valence/high arousal quad-
rants of the two-dimensional emotion space. Its mode
is major, which was predicted to evoke positive valence,
and the soft passages were predicted to evoke low arousal.
The soft and loud variation was predicted to produce
varying low and high levels of arousal. This selection is
henceforth referred to as ‘T’.
PROCEDURE
A tapping task was administered to establish beat-track-
ing ability. The task was designed to be a simplified par-
allel to the experimental task. It included a sounded and
an imagined condition where participants were required
to produce a steady tempo (160 bpm). In the sounded
condition, which served as practice, participants pressed
a keyboard key on a Yamaha PSR 500 digital keyboard
to synchronize with a sounded metronome clicking quar-
ter notes for 42 measures in standard 4/4 time. In the
main imagined condition, participants were given a two-
measure memory cue of the metronome tempo and then
asked to continue tapping for the next 40 measures in
the absence of the metronome clicks. Each trial consisted
of approximately 60 s of tap production. Cakewalk Pro
Audio 8 sound editing software was used to record par-
ticipants’ responses. Following the tapping task, partic-
ipants filled out a questionnaire that recorded age, gender,
and music training, as well as piece familiarity of the
three stimuli used in the experiment on a scale of 1–7.
Participants then were introduced to the experimen-
tal task and the Two-Dimensional Emotion Space (2DES)
software, a continuous response interface. We used the
Real Time Cognitive Response Recorder (RTCRR) devel-
oped by Schubert (2007). The program was presented
on a Macintosh G4 laptop computer. It allowed the user
to respond to x and y dimension variables by moving the
mouse along the continuum of the axes and was set to
record responses in real time at 0.5 s intervals. Valence
was presented on the x-axis and arousal on the y-axis.
The four quadrants of the interface as presented to par-
ticipants are illustrated in Figure 1: high valence/high
arousal, low valence/high arousal, low valence/low
arousal, and high valence/low arousal. The mouse began
each trial centrally located in the middle of the emotion
space. Valence was defined as the range of positive to neg-
ative emotions evoked by the music. Arousal was defined
as the range of emotions covering excited to sleepy. The
participants were then instructed as follows:
You are going to listen to a piece of music and your task
is to indicate the amount of valence and arousal
expressed by the music as it unfolds. Move the mouse
throughout the quadrants to indicate your response to
emotional output. You are rating the emotion you believe
the piece is trying to evoke.
Participants first executed practice trials with Vivaldi’s
Four Seasons: Spring (0:00–0:51). When participants
showed proficiency with the software and confirmed that
they understood the instructions they began the exper-
imental task. First was the sounded condition where par-
ticipants heard a musical excerpt and made valence and
arousal responses as the music unfolded. Two trials of
each of the three pieces were conducted in the sounded
condition. Next, in three trials of the imagined condi-
tion, participants heard a 5–8 s memory cue of the record-
ing and then responded as they imagined the rest of the
excerpt. To help participants keep their place in the music,
one-page reduced versions of the musical score were pre-
sented in both the sounded and imagined conditions
with the instructions that the score was supplemental
and did not need to be used if the participant thought it
unnecessary. The musical score was placed on the key-
board (directly below the emotion space) so participants
402
Brian J. Lucas, Emery Schubert, & Andrea R. Halpern
FIGURE 1. The RTCRR display.
Music2705_05 5/13/10 2:31 PM Page 402

could see both the software interface and the music. Some
participants chose to hold up the music so they were
looking at it alongside the screen. During the task the
experimenter was present in case any questions or prob-
lems arose.
Each trial of the sounded condition was conducted
before the participants began the imagined condition, to
allow the sounded condition to act as an additional train-
ing trial before the piece was to be imagined. For each
piece, approximately 15 min elapsed before the imagined
condition task was performed and participants were not
told that they would be imagining the same pieces for
the second task. This was done to reduce the likelihood
of participants focusing on memorizing their responses.
Order effects were addressed by counterbalancing the
presentation order of the three pieces across all partici-
pants. Following the sounded condition, another ques-
tionnaire was given, collecting information on a 1–7 scale
about the extent to which participants utilized the musi-
cal score and their perceived proficiency in imagining
the piece. Ratings on score use indicated moderate score
use in both Experiment 1 (M = 4.69) and Experiment 2
(M = 4.73). If time permitted, the experiment ended with
a general discussion to obtain information that could be
relevant in analysis. One participant, for example, dis-
cussed her role as a conductor and having performed the
Mozart piece.
Results
INCLUDED AND EXCLUDED TRIALS
With each of the 22 participants rating three excerpts twice
in the sounded condition and three times in the imagined
condition, a total of 330 continuous response profiles were
collected. In our analysis we used the second trial of the
sounded condition and the third trial of the imagined con-
dition. We chose to use the last trial of each condition
rather than an average of all the trials because this method
allowed the participants the most time to acquaint them-
selves with the task and the excerpts. Considering only the
last trial of the sounded and the imagined conditions, 132
trials were eligible for analysis, or 66 within-subject
sounded-to-imagined condition comparisons.
Next, we considered other factors that would require
a trial or a participant to be excluded from analysis. We
excluded response profiles from participants who scored
below a 3 (1–7 scale) on familiarity with an excerpt, or
who gave similarly low scores on the question,“How well
could you imagine this piece?” Many of these respon-
dents also showed incoherent, or “skywriting” patterns
with the mouse. A third problem arose from technical
difficulties or software error. These exclusions left a total
of 80 trials, or 40 sounded-to-imagined condition com-
parisons, for inclusion in our analysis.
TAPPING TASK
As a measure of rhythmic consistency, we chose to ana-
lyze intertap intervals. The average intertap interval
was 0.359 s (SD = 0.019), a 4.27% error (error range
4.00%–5.20%) from the actual interval of 0.375 s (at tempo
160 bpm). The consistency of tapping is in the range of
typical synchronization performance summarized by
Repp (2005), which indicates adequate rhythmic tapping
ability from all participants.
COMPARISON OF MEAN SOUNDED AND IMAGINED
SERIES FOR AROUSAL AND VALENCE
Comparisons of averaged sounded and imagined responses
for each response dimension across the three works were
calculated and plotted to provide initial visual comparison
of the tasks. The plots are shown in Figure 2a for arousal
and 2b for valence. In each case, a good similarity can be
seen between the imagined and sounded responses.
As a simple first pass on analysis, we conducted a
within-subject analysis by comparing the responses
(mouse positions) in the sounded condition to the
responses in the imagined condition of each participant.
An average of 85% of all individual arousal sounded-to-
imagined response profile correlations exceeded the crit-
ical value of r (as each of our music excerpts varied in
time, and thus number of responses collected, so too did
the degrees of freedom, and the corresponding critical
values.) For valence, 66% exceeded the critical value. We
save a more extensive and quantified analysis of these
results after presenting results of Experiment 2.
MUSICAL FEATURES ELICITING RESPONSES
Given that our second goal was to look at specific
sounded-imagined correspondences, we identified some
musical feature turning points that we predicted would
evoke an emotional response. They are characterized by
salient increases or decreases in the amount of a musical
feature (e.g., see Schubert, 2004), such as loudness, or a
change in the nature of a musical feature, such as a major
to minor change in mode. We then looked to see if imag-
ined and sounded condition profiles would show simi-
lar changes at those points.
An example of a musical turning point is given in
Figure 3, which shows a participant’s time-series data in
response to Beethoven. Emotional response value is on
the y-axis and it can be tracked by time in seconds along
the x-axis. One musical turning point we identified, at
25 s (measure 26), is characterized by a sudden decrease
in loudness and is labeled in the figure as B2.We examined
the 3 s window between B2 and B2’, where the apostrophe
Perception and Imagination of Emotion 403
Music2705_05 5/13/10 2:31 PM Page 403

(B2′) indicates that the response was expected to be due
to the feature change (B2). During the time between B2
and B2′, this participant responded with a clear decrease
in arousal in the sounded condition. For the imagined
condition the decrease in arousal from B2 to B2’ in the
time-series profile provides evidence that the turning
point was imagined.
Eight musical feature turning points were identified
for our analysis across all pieces. Seven of the events
involved loudness, pitch, note density, or a combination
of these features and were predicted to evoke changes in
arousal. One event involved mode and was predicted to
evoke a change in valence. Overall, participants responded
to the musical feature turning points in the sounded
music as expected. For all eight turning points, an aver-
age of 84% of the participants responded to the event in
the way we had predicted in the sounded condition.
Experiment 2
Experiment 2 was designed to complement Experiment
1 in two main respects. First, we explored emotional
labeling in Experiment 2. Although valence and arousal
have been used extensively in continuous response
experimentation, other labels, such as tension or pleas-
antness, have been used as well. There is some contro-
versy over which labels are the most appropriate and
provide the most clarity and accuracy in capturing emo-
tional experience (Schubert, 2001, 2010; Sloboda &
Lehmann, 2001).
We explored emotional labeling by adapting our soft-
ware to test for the dimension of emotionality.
Emotionality is measured in a single dimension and was
defined by Sloboda and Lehmann (2001) as “the capac-
ity of the performance at [any given] moment to suggest,
404
Brian J. Lucas, Emery Schubert, & Andrea R. Halpern
FIGURE 2. Mean sounded and imagined time series responses.
(a) Mean Arousal Response Time Series
for Beethoven (B), Mozart (M) and
Tchaikovsky (T).
FIGURE 3. A depiction of a musical feature turning point measurement for Beethoven.
(b) Mean Valence Response Time Series
for Beethoven (B), Mozart (M) and
Tchaikovsky (T).
Music2705_05 5/13/10 2:31 PM Page 404

communicate, or evoke musically relevant emotion.” Our
first goal was to determine if emotionality could be used
successfully in our continuous response paradigm. We
were also interested in whether using one dimension
might be easier for participants than tracking two dimen-
sions, due to the reduction in number of decisions at each
time point. Another hypothesis was that emotionality
responses would resemble those for arousal more so than
those for valence, as Schubert (2001) suggested.
Experiment 2 also allowed us to test whether or not
participants could replicate their continuous response
profiles (test-retest reliability). To do this, we retested
three participants from Experiment 1 on the valence and
arousal continuous response task. Successful replication
from these repeat participants would increase the relia-
bility of our Experiment 1 findings, augmenting the few
studies on test-retest reliability in continuous response
(Nagel et al., 2007; Schubert, 1999). Finally, we wanted to
apply cross-correlation analysis to the combined results
of both experiments, to capture the fine points of the
relationship between profiles generated in the sounded
and imagined conditions.
Method
PARTICIPANTS
The second experiment involved 11 Bucknell University
undergraduate students (18–22 years old; 8 female, 3
male). Seven participants were previously untested stu-
dents from the Introductory Psychology participant pool
and four participants were retested from Experiment 1. All
returning participants indicated on a questionnaire that
recollections of the task in Experiment 1 (15–18 weeks
prior) did not affect their responses in Experiment 2.
Music experience ranged from 8–14 years with an aver-
age of 11.27 years.
MATERIALS
The materials used in the tapping task and the continu-
ous response task were the same as in Experiment 1, with
one exception. One change was made to the interface of
the RTCRR for the continuous “emotionality” data col-
lection: Emotionality replaced valence along the x-axis,
and the y-axis was left unlabeled so that participants were
presented with a single response axis.
STIMULI
The three musical excerpts used in Experiment 1 also
were used in this experiment.
PROCEDURE
The first protocol was designed for repeat participants
from Experiment 1. Their first task involved continuous
response testing using the emotionality dimension.
Consistent with the methods of Experiment 1, partici-
pants responded to each excerpt twice in the sounded
condition and three times in the imagined condition. We
then retested these participants on the continuous
response task from Experiment 1 using the valence and
arousal dimensions. One trial was taken in each of the
sounded and imagined conditions in this task.
The second protocol was designed for previously
untested participants. This procedure was the same as in
Experiment 1 except that the experimental task meas-
ured emotionality in place of valence and arousal.
Results
The results of Experiment 2 are organized as follows.
First, we report which trials were retained and which
were excluded. This is followed by a pooling of data from
the two experiments in which a test-retest comparison
is reported, followed by a comparison of dimensions,
and finally a comparison of mean responses and indi-
vidual response lag structure between sounded and imag-
ined conditions.
EXCLUDED TRIALS
In the emotionality task, each of 11 participants rated
three excerpts twice in the sounded condition and three
times in the imagined condition, for a total of 165 con-
tinuous response profiles. We used the second trial of the
sounded condition and the third trial of the imagined
condition in our analysis of the emotionality task. This
left 66 trials eligible for analysis, or 33 within-subject
sounded-to-imagined condition comparisons.
For each retest participant (n = 3; one person’s retest
data file had technical problems, so data for three par-
ticipants remained) on the valence-arousal task, we col-
lected one trial for each excerpt in both the sounded
and imagined conditions. This yielded 18 trials for our
analysis.
For the emotionality tasks, no participants were
excluded from analysis and, other than the trials excluded
because of our trial selection method, no individual tri-
als were excluded. All participants scored a 3 or above
(using a 1–7 scale) on excerpt familiarity and perceived
proficiency in imagining the excerpt.
All participants completed the tapping task (either in
Experiment 1 or 2) sufficiently well to be retained for the
study, using similar criteria for new participants in
Experiment 2 as were used in Experiment 1.
TEST-RETEST COMPARISON
Three participants completed the arousal and valence
response tasks twice, allowing us to examine test-retest
reliability in a small sample. Repeated measures were
Perception and Imagination of Emotion 405
Music2705_05 5/13/10 2:31 PM Page 405

compared for each of the three participants by piece (3),
dimension (arousal and valence) and condition (sounded
and imagined) using a Pearson correlation analysis. As
shown in Table 1, all participants’ repeated responses
most closely resembled the original for the arousal-
sounded (AS) condition, regardless of the piece (all cor-
relations were significant). Further, all participants’
repeated responses were significantly correlated for the
Tchaikovsky regardless of the condition. The Beethoven
responses were least reliable for participant 1, as reflected
by the negative correlations (which were due to the
numerous changes in response that were “out of phase”
when the valence-sounded (VS) condition in particular
[r =−0.50] was repeated).
Test-retest analyses showed that valence responses
had lower reliability than arousal responses, and the
responses in the imagined condition were less reliable
than the sounded conditions. The pauses and impulsive
nature of the Beethoven example may have been respon-
sible for the less consistent responses found compared
to the other pieces. However, seven of the nine arousal-
imagined (AI) correlations were significant, many
impressively so—and six of the nine valence-imagined
(VI) responses—suggesting that reliability of imagined
responses, although lower than sounded responses, is
still considerable.
VISUAL COMPARISON OF SOUNDED AND IMAGINED
SERIES ON EMOTIONALITY JUDGMENTS
The plots of the superimposed time series comparing
sounded and imagined conditions for the emotional-
ity dimension is shown in Figure 4. Visual inspection
of this figure shows similarity between sounded and
imagined conditions at least as striking as we saw in
Experiment 1. All three pieces showed quite a large range
of emotionality judgments in both conditions, which
is more similar to the arousal than valence profiles seen
in Experiment 1. We quantify these relationships in the
next sections.
ARE AROUSAL AND EMOTION DIMENSIONS RELATED IN
SOUNDED MUSIC?
CCFs (Cross-Correlation Functions, see Campbell, Lo,
& MacKinlay, 1997) were performed between pairs of
emotion response dimension mean time series for the
sounded conditions to determine whether any pairs had
similarities. Cross-correlation takes two time series and
performs several correlation analyses between them. Each
correlation analysis is performed at different time lags
between the two series. This allows identification of the
lag at which the two series are maximally correlated.
Further, similarity is indicated by large peaks in correla-
tions for at least one lag in the CCF. The CCFs in Figure 5
show that arousal and emotion produce the strongest
and most frequent correlation coefficients (correlation
coefficients above the confidence interval), with the other
two pairs (emotion with valence and arousal with valence)
less so, confirming the prediction that emotion and
arousal tap into a similar semantic dimension.
ANALYSIS OF LAG STRUCTURE BETWEEN IMAGINED
AND SOUNDED CONDITIONS
In analyzing cross-correlations between imagined and
sounded responses for each of the three tasks, two analy-
ses were of interest: (1) The number of responses that
produced significant lags for at least one lag of the CCF.
A large number of responses producing at least some sig-
nificant lag for a condition indicates that participants are
able to reproduce the imagined condition in a way that
resembles the sounded condition response. (2) Of these
“significant peak lag”responses, the lag at which the peak
correlation occurred within the comparison. This will
provide information about the lag structure between
imagined and sounded responses. For example, is there
a greater delay in imagined responses when arousal is the
dimension being responded to? If so, how long is this
delay (lag)?
For these analyses, a difference transformation was
made. This means that instead of examining the absolute
values recorded by participants, the change in value from
one sample to the next is computed. This approach has
been shown to reduce the effects of serial correlation and
thus provide more valid results (see Schubert, 2002).
Further, only the 100 samples beginning from the 8th
second were used in the analyses. This was to ensure that
406
Brian J. Lucas, Emery Schubert, & Andrea R. Halpern
FIGURE 4. Mean emotionality response time series for Beethoven (B),
Mozart (M), and Tchaikovsky (T).
Music2705_05 5/13/10 2:31 PM Page 406

results were not biased by the presence of sounded music
in the sounded condition and the imagined condition
(in which the music was used for cueing the participant
for the first 5 to 7 s). Sampling was made at 2Hz.
The CCF output was analyzed for presence of at least
one significant correlation at any lag, and for these, the
lag at which the peak significant correlation occurred.
This was determined by categorizing a time series
response pair (imagined versus sounded) as “significant”
if the largest correlation coefficient was above the con-
fidence interval, which in this case was set to one standard
error (using the standard error of cross-correlation at
each lag as reported by Box & Jenkins, 1976, p. 376).
That is, if for an imagined-sounded comparison, no cor-
relations were greater than 1 SE (at any lag in the CCF),
then the pair was categorized as “not significant.” Table
2 summarizes the results of the analyses. The mean lag
between imagined and sounded conditions was less than
2 samples (1 s) for each cell (in Table 1), with the excep-
tion of emotionality response in Beethoven, in which
the mean peak imagined response leads the sounded
response by 2.73 (1.36 s). This means that imagined
Perception and Imagination of Emotion 407
FIGURE 5. Analysis of lag structure between emotional dimensions in sounded music.
(a) Mean Arousal versus Mean Emotionality (b) Mean Arousal versus Mean Valence
(c) Mean Emotionality versus Mean Valence
Music2705_05 5/13/10 2:31 PM Page 407

responses tended to be slightly rushed with respect to
the sounded condition when rating the emotion
expressed for the Beethoven excerpt. However, it also
should be noted that this mean peak lag returned a rel-
atively large variability in location (second only to
Mozart arousal), with peak lags distributed with a SD
of 5.14 samples (2.57 s) across participants.
The most consistent responses (considering peak devi-
ation) were for the valence response for Beethoven (SD
= 1.54 samples), and the arousal response to Tchaikovsky
(SD = 1.98 samples). The closest-to-instantaneous mean
lag structure (i.e., imagined and sounded condition
response timings best matched) occurred for arousal
responses for all three pieces, in each case less than ±0.36
samples (±0.18 s). However, the variability of the peak
lag for Beethoven and Mozart are relatively large (3.14
samples and 5.80 samples respectively). This suggests
that participants had less idiosyncratic recall of imag-
ined arousal responses for the Tchaikovsky. Tchaikovsky
was also the only piece where the average peak lag was
leading (imagined condition response occurring faster rel-
ative to the sounded condition—see negative coefficients
in third column of Table 2). The results indicate that lag
structure between sounded and imagined responses are
quite reliably correlated, with 83% to 100% of participants
producing at least one significant correlation (at some
408
Brian J. Lucas, Emery Schubert, & Andrea R. Halpern
TABLE 1. Pearson Correlation Coefficients of Responses For Repeat Participants in Experiments 1 and 2.
Condition
Stimulus Participant VS VI AS AI Mean
B1 −
.50 −.11 .39 −.02 −.05
2 .86 .02
.78 .77 .61
3 −
.38 −.59 .78 .85 .17
Mean .00 −.23 .65 .55
M1 −
.12 .41 .84 −.10 .34
2 .59 .86 .96 .89 .83
3 .77 .78 .92 .84 .83
Mean .49 .69 .91 .57
T 1 .81 .83 .77 .82 .81
2 .95 .92 .95 .98 .95
3 .96 .88 .97 .89 .93
Mean .91 .88 .90 .90
Note. Underlined correlation coefficients are either negative or non-significant (p = .05). Bold indicates all coefficients in the row or column were
significant. Df varies for each cell, see text. Codes: V = Valence, A = Arousal, S = Sounded, I = Imagined, B = Beethoven, M = Mozart, T = Tch ai ko vsk y.
TABLE 2. Peak Lag Statistics by Stimulus and Condition.
Mean Peak Lag SD Peak Lag % Significant
Stimulus Task # Responses (in samples) (in samples) (>1 SE)
BA16 −.03 3.14 94
V 16 .03 1.54 94
E12 −2.73 5.14 92
M A 12 .03 5.80 100
V 12 2.00 4.13 83
E 12 .55 3.15 92
TA15 −.04 1.98 93
V15 −.71 3.42 93
E 12 1.36 3.49 92
Note. B = Beethoven, M = Mozart, T = Tchaikovsky, V = Valence, A = Arousal, E = Emotionality. % Significant refers to the percentage of participants for
whom the correlation coefficient of the peak lag between imagined and sounded conditions was greater than the significance level of 1 SE.
Music2705_05 5/13/10 2:31 PM Page 408

lag) for each of the three pieces, and each of the three
conditions.
To exemplify these results, consider the case of the
valence response from participant 21 to the Mozart.
Figure 6a demonstrates that the imagined valence
response time series resembles the sounded valence
series, but the imagined responses occur relatively ear-
lier (most clearly depicted between the 44th and 53rd
second of the response). This difference is quantified
in the CCF plot (Figure 6b), where a peak at −7 (3.5 s)
samples is clearly visible. It tells us that the imagined
response is similar to the sounded response but lead-
ing (rushing or jumping ahead) by 3.5 s. It is these peak
correlation coefficients that were collated and reported
in Table 2.
General Discussion
This paper described two experiments that investigated
how well the emotion associated with musical excerpts
could be recalled. The novel aspects of the study were
that emotional responses were recalled continuously as
music unfolded mentally, and that three emotional
dimensions were measured. Further, one-minute excerpts
of late Classical and Romantic music were used to allow
a variety of emotions within works that were familiar
and did not refer to words.
At the most basic level, this study is one of a number
of studies that report evidence of use of auditory
imagery (Halpern, 1988a; Hubbard & Stoeckig, 1992;
Intons-Peterson, 1992). The consistency of profiles over
sounded and imagined conditions demonstrates that a
majority of participants could successfully track emo-
tion in a sounded excerpt of music, and that they could
then extract emotion from their memory of that
excerpt.
It is evident that familiarity with the piece is critical
for this complex task. We screened all of our participants
for music training, which made it likely they would have
encountered famous pieces like these. Furthermore, we
insured that they knew these particular pieces. However,
we think it unlikely that our trained participants had
already coded the emotional changes in each piece that
we required them to externalize in the moment-to-
moment paradigm of continuous response. Rather, we
propose that participants recalled the requested piece
and made the requested judgments de novo, utilizing the
memory representations. Combined with similarity in
response to features, this success suggests that the mem-
ory representation is quite detailed. The representation
is also apparently quite stable, as evidenced by the simi-
lar responses over a lengthy period by our participants
who took part in both experiments.
One argument against this notion is that emotional
responses were consciously utilized during the sounded
condition and that they were therefore readily accessible
in the auditory image of the piece, within a session.
However, we think this is unlikely, as the three excerpts
were tested back-to-back in the sounded condition before
being presented in the imagined condition. Participants
were not aware that they would be responding to the same
piece, and there was no reason to believe that they used
conscious strategies to memorize responses. It also strains
credulity that the precise judgments could be replicated
Perception and Imagination of Emotion 409
FIGURE 6. Time series and CCF for participant 21 valence response to Mozart.
(a) Differenced Time Series of Valence Imagined
and Valence Sounded
(b) CCF of the Two Series Shown in (a), Sounded
versus Imagined
Music2705_05 5/13/10 2:31 PM Page 409

on demand, at the sampling rate (2 Hz) used for contin-
uous responses.
We think it particularly noteworthy that the emotions
extracted from the musical image are similar to the emo-
tion expressed by sounded music. This similarity allows
us to propose that the underlying processing mechanisms
for the two conditions are similar, as has been suggested
for musical imagery tasks involving pitch, time, or timbre
judgments (Halpern, 1988a, 1988b; Halpern et al., 2004).
The retrieval of auditory images for music thus may not
only be fairly easy for listeners, but also enjoyable (many
people report being able to retrieve images of their favorite
tunes, in whatever genres they are familiar with).
In this study, lag structure was more or less instanta-
neous (typically less than 0.5 s), meaning that partici-
pants did not fluctuate in time (slow down or speed up)
when reporting emotion expressed from a musical image
compared to a sounded recording. Thus, although
imagery processes are typically slow compared to other
kinds of memory retrievals (judgments are on the order
of seconds in standard mental rotation or other imagery
tasks, as compared to, for instance, a few hundred mil-
liseconds in a lexical decision task), in this task, the emo-
tional judgments apparently were available at about the
same time as the musical information itself was retrieved
(on the order of less than one second). It may be the case
that using very familiar pieces allowed some anticipa-
tion of the next note or group of notes (Leaver, Van Lare,
Zielinksi, Halpern, & Rauschecker, 2009). The only occa-
sion when there was notable variability was for the piece
that had several stop-starts (fermata), namely the open-
ing of Beethoven’s Fifth Symphony. This could be con-
nected with the mismatch between perceptions of silences
(Fabian & Schubert, 2008; ten Hoopen, et al., 2006) and
the subsequent variability in response that this might
have caused.
However, the most important evidence that emotional
identification in sounded and imagined conditions was
the same is through error analysis. If we set the baseline
measurement error from the tapping task (which was,
on average, 4.27%), then we would expect an error
through tempo instability alone to produce a margin of
2.14 s (4.27% of the 50 s excerpt) or 4.28 samples. From
Table 2 we can see that of none of the nine experimen-
tal cells (three pieces by three emotion dimension) had
a mean lag structure greater than this value (greatest
being −2.73 samples for emotionality identification in
Beethoven). While the range of lag structure is fairly large
for some pieces (in particular, Beethoven emotionality,
as discussed), these findings nevertheless suggest that
timing may be better overall when tracking emotion in
familiar music than it is for a simple tapping task. The
“high-level” cognitive task of tracking emotion in famil-
iar music may facilitate low-level tempo tracking because
tempo is integrated into the recall of the higher-level per-
cept. This could also be indicative of the greater diffi-
culty in the production task (tapping a tempo) versus a
perception or emotion task. Further research on whether
the nature of the task (emotion rating versus pure tempo
tapping; recall versus production) may influence tempo
is a corollary of our study.
The relationships among valence, arousal, and emo-
tionality were consistent with that predicted by Schubert
(2001, 2010), supporting the idea that emotionality and
arousal may be semantic constructs with shared ontology.
However, imagined emotionality did have larger mean
lag than arousal (and valence, for that matter), and tended
to rush ahead with respect to emotionality response in the
sounded condition. This is in contrast to imagined
arousal, which remained instantaneous with sounded
arousal. It is, therefore, possible that “arousal” provides
a more reliable and stable indication of emotional
response than its related counterpart, emotionality. This
paper does not aim to address the relationship and fur-
ther research may explain the similarity in profiles
between arousal and emotionality, but at the same time
the difference in lag structure between the two.
In summary, it appears that despite the attention-
demanding nature of musical imagery, people can reli-
ably maintain that representation and extract novel
information from it. We also note that the length of time
in which we asked our participants to do these tasks was
much longer than in most prior auditory imagery exper-
iments, suggesting that these are not fleeting experiences.
And we showed that the emotional information thus
extracted is similar to the perceptual experience in impor-
tant ways. In this context, a quote from a musician
entombed for 18 hours in the rubble of the Haiti earth-
quake of 2010 is illuminating. He passed the time by
imagining some of his familiar concertos:“For example,
if I perform the Franck sonata, which is [sic] 35 minutes
long in my honors recital at Juilliard, then I would bring
myself to that time. That allows me not only to kill time,
but also to mentally take myself out of the space where I
was” (National Public Radio, 2010, italics added).
Author Note
Correspondence concerning this article should be addressed
to Andrea Halpern, Psychology Department, Bucknell
University, Lewisburg, PA 17837.
E-MAIL: ahalpern@
bucknell.edu
410
Brian J. Lucas, Emery Schubert, & Andrea R. Halpern
Music2705_05 5/13/10 2:31 PM Page 410

Perception and Imagination of Emotion 411
References
BAILES, F. (2007). The prevalence and nature of imagined
music in the everyday lives of music students. Psychology of
Music, 35, 555–570.
B
OX,G.E.P.,& G.M.JENKINS. (1976). Time series analysis:
Forecasting and control. San Francisco, CA: Holden-Day.
C
AMPBELL,J.Y.,LO,A.W.,& MACKINLAY, A. C. (1997). The
econometrics of financial markets. Princeton, NJ: Princeton
University Press.
E
VANS,P.,& SCHUBERT, E. (2008). Relationships between
expressed and felt emotions in music. Musicae Scientiae, 12,
75–99.
F
ABIAN,D.,& SCHUBERT, E. (2008). Musical character and
the performance and perception of dotting, articulation
and tempo in 34 recordings of Variation 7 from JS Bach’s
Goldberg Variations (BWV 988). Musicae Scientiae, 12,
177–206.
G
ABRIELSSON, A. (2010). Strong experiences with music. In P. N.
Juslin & J. A. Sloboda (Eds.), Handbook of music and emotion:
Theory, research, applications. (pp. 547–574). Oxford: Oxford
University Press.
G
ABRIELSSON, A., & LINDSTROM, E. (2001). The influence of
musical structure on emotional expression. In P. N. Juslin &
J. A. Sloboda (Eds.), Music and emotion: Theory and research
(pp. 223–248). Oxford: Oxford University Press.
G
AGNON, L., & PERETZ, I. (2003). Mode and tempo relative
contributions to “happy-sad” judgments in equitone
melodies. Cognition and Emotion, 17, 25–40.
H
ALPERN, A. R. (1988a). Mental scanning in auditory imagery
for songs. Journal of Experimental Psychology: Learning,
Memory, and Cognition, 14, 434–443.
H
ALPERN, A. R. (1988b). Perceived and imagined tempos of
familiar songs. Music Perception, 6, 193–202.
H
ALPERN, A. R., ZATORRE, R., BOUFFARD, M., & JOHNSON,
J. (2004). Behavioral and neural correlates of perceived
and imagined musical timbre. Neuropsychologia, 42, 1281–
1292.
H
EVNER, K. (1935). The affective character of the major and
minor modes in music. American Journal of Psychology, 47,
103–118.
H
UBBARD, T. L., & STOECKIG, K. (1992). The representation
of pitch in musical imagery. In D. Reisberg (Ed.), Auditory
imagery (pp. 199-235). Hillsdale, NJ: Lawrence Erlbaum
Associates, Inc.
I
NTONS-PETERSON, M. J. (1992). Components of auditory
imagery. In D. Reisberg (Ed.), Auditory imagery (pp. 45–71).
Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.
K
OZHEVNIKOV, M., KOSSLYN, S., & SHEPARD, J. (2005). Spatial
versus object visualizers: A new characterization of visual
cognitive style. Memory and Cognition, 33, 710–726.
L
EAVER, A. M., VAN LARE,J.,ZIELINSKI,B.,HALPERN, A. R.,
& R
AUSCHECKER, J. P. (2009). Brain activation during
anticipation of sound sequences. Journal of Neuroscience, 29,
2477–2485.
N
AGEL,F.,KOPIEZ, R., GREWE,O.,& ALTENMULLER,E.
(2007). EMuJoy: Software for continuous measurement of
perceived emotions in music. Behavior Research Methods,
39, 283–290.
N
ATIONAL PUBLIC RADIO (2010, January 23). Wife, school lost
in quake, violinist vows to rebuild [Radio program]. Retrieved
from http://www.npr.org/templates/story/story.php?storyId=
122900781&ft=1&f=122900781
R
EPP, B. H. (2005). Sensorimotor synchronization: A review of
the tapping literature. Psychonomic Bulletin and Review, 12,
969–992.
R
USSELL, J. A. (1980). A circumplex model of affect. Journal of
Social Psychology, 39, 1161–1178.
S
CHUBERT, E. (1999). Measuring emotion continuously:
Validity and reliability of the two-dimensional emotion-
space. Australian Journal of Psychology, 51, 154–165.
S
CHUBERT, E. (2001). Continuous measurement of self-report
emotional response in music. In P. N. Juslin & J. A. Sloboda
(Eds.), Music and emotion: Theory and research (pp. 393–414).
Oxford: Oxford University Press.
S
CHUBERT, E. (2002). Correlation analysis of continuous
emotional response to music: Correcting for the effects
of serial correlation. Musicae Scientiae, Special Issue, 213–
236.
S
CHUBERT, E. (2004). Modeling perceived emotion with
continuous musical features. Music Perception, 21,
561–585.
S
CHUBERT, E. (2007). Real time cognitive response recording.
Proceedings of the Inaugural International Conference on Music
Communication Science, Sydney, Australia. Paper retrieved
from http://marcs.uws.edu.au/links/ICoMusic/ArchiveCD/
fullpaper.html
S
CHUBERT, E. (2010). Continuous self-report methods. In P. N.
Juslin & J. A. Sloboda (Eds.), Handbook of music and emotion:
Theory, research, applications (pp. 223–253). Oxford: Oxford
University Press.
S
CHUBERT, E., & DUNSMUIR, W. (1999). Regression modeling
continuous data in music psychology. In S. W. Yi (Ed), Music,
mind, and science (pp. 298–352). Seoul, South Korea: Seoul
National University.
S
CHUBERT, E., EVANS,P.,& RINK, J. (2006). Emotion in real
and imagined music: Same or different? In M. Baroni, A. R.
Addessi, R. Caterina, & M. Costa (Eds.), Proceedings of the
Ninth International Conference on Music Perception and
Cognition (pp. 810–814). Bologna, Italy.
Music2705_05 5/13/10 2:31 PM Page 411

SERAFINE,M.L.,CROWDER, R. G., & REPP, B. H. (1984).
Integration of melody and text in memory for songs. Cognition,
16, 285–303.
S
LOBODA, J. A., & LEHMANN, A. C. (2001). Tracking perform-
ance correlates of changes in perceived intensity of emotion
during different interpretations of a Chopin piano prelude.
Music Perception, 19, 87–120.
S
WELLER, J. (1988). Cognitive load during problem solving:
Effects on learning. Cognitive Science, 12, 257–285.
S
WELLER, J. (2006). Discussion of ‘emerging topics in cognitive
load research: Using learner and information characteristics
in the design of powerful learning environments.’ Applied
Cognitive Psychology, 20, 353–357.
TEN HOOPEN, G., SASAKI,T.,NAKAJIMA,Y.,REMIJN, G.,
M
ASSIER,B.,RHEBERGEN, K. S., & HOLLEMAN, W. (2006).
Time-shrinking and categorical temporal ratio perception:
Evidence for a 1:1 temporal category. Music Perception,
24, 1–22.
412 Brian J. Lucas, Emery Schubert, & Andrea R. Halpern
Music2705_05 5/13/10 2:31 PM Page 412