Music2702_02 11/16/09 6:00 PM Page 89
Absolute Pitch Ability 89
A D ISTRIBUTION OF
BY
PATRICK B ERMUDEZ AND R OBERT J. Z ATORRE
Montreal Neurological Institute, McGill University,
Montreal, Canada & BRAMS Laboratory, Montreal,
Canada
BEHAVIORAL ASSESSMENT OF ABSOLUTE PITCH ( AP )
ability over the better part of the last century has strongly
suggested that a variety of proficiency levels exists and
can only be more comprehensively described with the
use of rigorous testing providing precise and unbiased
reaction times for all responses. This study describes the
design, implementation and validation of a computerized
test of absolute pitch and resulting data for 51 musicians,
27 of whom self-reported as AP possessors. The test was
sensitive to previously reported differences in accuracy
and timing for C major diatonic versus non-diatonic
notes and showed a range of performance, from perfect
to random, including a substantial number of interme-
diate levels of proficiency. We discuss the implications
of detecting such a distribution of behavior as well as
the effect of test design and scoring strategies on that
distribution.
Received November 28, 2008, accepted June 24, 2009.
Key words: absolute pitch, behavior, computerized test,
distribution, scoring
A
BSOLUTE PITCH (AP) HAS LONG BEEN OF INTEREST
to scientists studying music cognition (Bachem,
1937; Ward, 1999). Simply and typically defined,
it is the ability to identify the names of musical pitches
without reference to a standard (Bachem, 1937; Ward,
1999). Part of its scientific attraction stems from the
fact that it is a rare ability whose overt manifestation
requires complex cognitive function, and yet it is read-
ily amenable to operationalization. The reasons for the
uncommon occurrence of AP are ill-understood and
the necessary and sufficient conditions for its develop-
ment remain the subject of considerable speculation
and investigation (Baharloo, Service, Risch, Gitschier, &
Freimer, 2000; Gregersen, Kowalsky, Kohn, & Marvin,
2000; Levitin & Rogers, 2005; Miyazaki, 2004; Ross,
Music Perception VOLUME 27, ISSUE 2, PP. 89101, ISSN
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.
A BSOLUTE P ITCH A BILITY AS R EVEALED
C OMPUTERIZED T ESTING
Gore, & Marks, 2005; Zatorre, 2003). The strongest
forms of the ability are far from universal and not all
those who are exposed to equivalent musical tuition (to
the extent that this can ever be determined) develop the
ability. Sufficiently early and intensive music training
seems to be an important factor but it likely interacts
with certain biological/cognitive dispositions and nor-
mal developmental stage in forming propitious circum-
stances that favor its emergence.
To date, there exists no standard method for the
behavioral determination of absolute pitch ability, and
many studies have simply relied on participant self-
report. Among those studies that have verified AP
explicitly, a variety of approaches have been espoused,
most having required the identification (verbal, writ-
ten, or other) of a given tone in the absence of a refer-
ence tone (Takeuchi & Hulse, 1993; Ward, 1999).
Stimuli have most commonly been piano notes (real
and synthesized), sine waves, and multiharmonic syn-
thetic tones, and there has been little consistency in
terms of administration and analysis parameters such
as stimulus length, number of trials, reaction time col-
lection, scoring strategy, performance criteria, etc. As
there seems to be a gamut of AP ability, ranging from
excellent performance with pure tones over a wide fre-
quency spectrum (showing ability that is least stimulus
bound, perhaps corresponding to Bachems genuine
AP; Bachem, 1937) to moderate performance within a
more narrow range of notes from a preferred instru-
ment, it is most desirable to have control over stimulus
type, accuracy and response latency as possible means
for distinguishing different subtypes of AP (Baharloo,
Johnston, Service, Gitschier, & Freimer, 1998; Miyazaki,
1989, 1990; Takeuchi & Hulse, 1993). The use of piano
tones alone risks what Ward and Burns (1982) have
referred to as the testing of absolute piano, meaning
generally that timbre and other cues characteristic of an
instrument can be used in addition to fundamental fre-
quency to help identify notes. Although such a phe-
nomenon is in itself interesting, it may be best
considered a more stimulus-bound variety of absolute
pitch. Probable differences in performance and strategy
among AP musicians have been recognized for some
time (Bachem, 1937) yet limited efforts have been made
0730-7829, ELECTRONIC ISSN 1533-8312 2009 BY THE REGENTS OF THE UNIVERSITY OF CALIFORNIA . ALL
DOI:10.1525/MP.2009.27.2.89
Music2702_02 11/16/09 6:00 PM Page 90
90 Patrick Bermudez & Robert J. Zatorre
to describe and quantify this variability and take it into
account in experimental contexts. It seems therefore
that relying on participants self-report may only be
adequate during initial stages of recruitment.
As alluded to above, choice of stimulus, administra-
tion parameters, and scoring strategies all have con-
tributed to considerable variability among reported
results (Miyazaki, 1989; Takeuchi & Hulse, 1993).
Relatively more discussion has been dedicated to possi-
ble effects of stimulus type than to administration
details; specifically the importance of collecting accu-
rate reaction times and the possible effects of various
scoring criteria on the perceived grouping of partici-
pants performance. Response latency is one dimension
of performance that holds the potential to resolve par-
ticipants who are indistinguishable from each other
solely on the basis of a percent correct or other similar
score. As has been observed by other researchers
(Miyazaki, 1989, 1990; Takeuchi & Hulse, 1991), we
might expect that participants with a weaker form of
AP or Bachem-style pseudo-AP ultimately could
achieve far better than chance performance but would
require more time to respond on average, likely due to
the use of alternative strategies including relative pitch
judgements from a limited number of internalized
pitch class representations (e.g., a stable memory for a
frequent tuning note such as A = 440 Hz).
By far the most common scoring of a typical test of
absolute pitch, where a participant is presented a note
and asked for a verbal, written or other identification of
that note, has been to assign a point for an exactly cor-
rect chroma identification and, in fewer studies, a
point, 3/4-point, or 1/2-point for semitone deviations.
Given that there is no community-wide operational
definition or scoring convention and often no clear dis-
continuity between purported sub-populations among
musicians, these are necessarily arbitrary criteria. One
might reasonably conjecture that the stricter of the
approaches, where only exact responses are considered,
risks ignoring relatively high performers who consis-
tently approximate the correct answer but are not quite
resolving the semitone, and therefore not being awarded
the crucial points, or participants who perform well but
with a consistent shift of a semitone or more due to
their learning history (e.g., with a mistuned instru-
ment) or some paracusis. An information transfer
approach (Miller, 1956; Pollack, 1952; Ward & Burns,
1982) that can recover the consistency of information
transmitted by the participant (expressed in terms of
bits of information) is somewhat appropriate for dealing
with the latter and a simple measure of average deviation
sufficient for the former. Examining the distribution of
a participants responses and a measure of average devi-
ation also provides intuitively accessible estimates of
the average resolution of responses; that is, the ability
to achieve a sufficiently narrow pitch-class distinction
so as to allow reliable recognition and labelling at the
semitone level. Of further concern is the possibility
that the strictest and traditional scoring method (only
crediting exactly correct responses) may exaggerate
the occasionally described bimodal distribution of AP
performance among musicians. In other words, it is
sometimes suggested (e.g. Athos et al., 2007; Carroll,
1975; Miyazaki, 1988) that there are two clearly resolv-
able populations, those who can and those who cannot
perform an AP task, with very little (if any) overlap
between the two distributions. If one ignores large parts
of the variance in performance by only considering
exactly correct responses (or even 1 semitone), one
can easily imagine the effect on a pair of gamma-like
distributions; namely, that they may be rendered more
narrow, with artificially sharp boundaries thereby
accentuating the bimodal impression. Assigning scores
to semitone errors serves to palliate this problem some-
what but also unnecessarily blurs a potentially fine
distinction in ability between the participant who
performs perfectly and the one who performs with a
lower-resolution in a two semitone band centered on
the correct answer. As suggested in the previous para-
graph, this information is recovered with a trivially
simple measure of mean deviation from the correct
response.
Athos et al. (2007) used an Internet-based test and
survey to collect data from well over 2000 self-selected
participants, 44% of whom were categorized as being
possessors of a particularly strong form of AP. Their
distribution of scores indicates some degree of bimodal-
ity, suggesting that there are two largely distinct popu-
lations accounting for the worst and best performers.
The authors drew two conclusions based on this distri-
bution: that their data resolve the question of whether
AP participants lie at the extreme of a continuum of
ability or form a distinct population (clearly the latter,
in their opinion), and that AP could have its origins in
one or a few genes. Additionally, they document a finding
consistent with the anecdotally described age-related
shift in perception that leads some participants to
perceive notes as sharper than they once did (Ward,
1999). They also speculate on a possible perceptual
magnet effect for the pitch class A, following mainly
from the finding that the class G#, in particular, often is
mislabelled as A (which we read as consistent with
some of the results obtained by Van Krevelen, 1951,
who recorded mean recognition frequencies for G# and
Music2702_02 11/16/09 6:00 PM Page 91
A# that were higher and lower than standard, respec-
tively). The study highlighted the value of web-based
data collection as a technique complementary to those
normally used in this area of investigation. Several
notes of caution should be voiced, however. Any claims
about the population must be constrained and tempered
by knowledge of the limitations inherent to the sampling
methodology used, which are essentially those of most
unconstrained survey techniques (Kaplan & Saccuzzo,
1993). For instance, participants were self-selected and
motivated to participate out of personal interest, from
which one might anticipate an over-representation of
AP possessors given the advertised aims of the study
(as suggested by the conspicuously high 44% of respon-
dents showing particularly strong AP-1 performance),
thereby accentuating one of the two supposed modes.
Secondly, the scoring method used, where exactly correct
answers were awarded one point and semitone devia-
tions 3/4 of a point (0 for all other responses including
any requiring more than 4 seconds), risks exaggerating
the separation between the best and worst performers
by discarding intermediate performance. Given these
and other related issues, it seems imprudent to declare
the matter resolved. In what concerns the genetic claim,
even if AP is reduced to its most bare and requisite fea-
tures, there is no necessity that it should be attributed
to a single gene. There are examples of traits and diseases
with simple genetic origins yet complex manifestations,
but also of simple traits being governed by complex
and large collections of genes (NLM, 2008). More
fundamentally, a bimodal distribution of measured
behavior does not necessarily map to simple underly-
ing biology and could just as easily suggest a poly-
genic or multifactorial inheritance model in which an
interaction of several genes and environment conspire
to produce a variety of phenotypic manifestations
(Benjamin, 2008), i.e., levels of proficiency. Put another
way, the effect we are measuring could be one among
many manifestations of more general underlying factors
(genetic/biological/cognitive). Indeed, there is a line of
parsimonious reasoning that informs us of the unlike-
lihood of this ability being so completely isolated from
the rest of brain function, perfectly self-encapsulated
and having no relation whatsoever to any other aspects
of cognition (Bermudez, 2008; Bermudez & Zatorre,
2009; Chin, 2003; Costa-Giomi, Gilmour, Siddell, &
Lefebvre, 2001; Moreno Sala & Costa-Giomi, 2008). We
might also observe, due to a metric of limited dimen-
sionality and resolution, similar behavior accomplished
by means of different neural infrastructure that could,
in turn, be attributable to different biological/genetic
factors.
Absolute Pitch Ability 91
Certain predictions follow from AP literature. We
obviously expect significant differences between self-
reported AP and non-AP (NAP) possessors in various
measures of accuracy and reaction time for chroma
identification (higher and shorter, respectively, in the AP
group). We also expect that diatonic notes of the C
major scale will be identified more accurately and more
quickly than the accidentals of C major (Miyazaki, 1990;
Takeuchi & Hulse, 1991). Additionally, intermediate
performers might show different response patterns and
reaction times as compared to either high-performing
AP or randomly performing NAP musicians, perhaps
reflecting alternative strategies and/or weaker, lower res-
olution chroma representations. This paper presents a
detailed description of a computerized test of absolute
pitch, data collected to date, and future directions. We
also will describe the reliability of the test and explore
the advantages of collecting precise reaction times as
well as alternative descriptions of performance that do
not rely on traditional scoring methods.
Method
Participants
Participants primarily were recruited through the music
faculties of two local universities. The posted advertise-
ments simply stated that we were seeking musicians both
with and without absolute pitch, with at least 10 years
of training/practice to participate in behavioral testing
and a neuroimaging experiment. All participants gave
informed consent and all procedures were approved
by the Research Ethics Committee of the Montreal
Neurological Institute, McGill University. Sample con-
stitution was as follows: 51 musicians, 39 females and
12 males, 27 of which self-reported as AP possessors.
They had an average age of 23.1 years (SE = 0.52), average
age of commencement of playing and/or training of
6.1 years (SE = 0.32), and average of 16.4 years of total
playing and training experience (SE = 0.63).
Stimuli
Fundamental frequencies of tones corresponded to
notes ranging from C3 to B5 (three octaves) and were
based on an A = 440 Hz equal temperament tuning,
with a 21/12:1 frequency ratio between adjacent semi-
tones. Each note was presented three times, once at
each of three different intensities (-1, -4 and -7 dB) in
order to eliminate the use of intensity cues to aid judge-
ments. This summed to 108 trials. The notes were syn-
thetic tones generated by summing a series of sinusoid
Music2702_02 11/16/09 6:00 PM Page 92
92 Patrick Bermudez & Robert J. Zatorre
(a) (b)
C C# b
B D C -> B
Bb D equidistant from the trial starting point which, together
A#
A D#
Eb
Ab
E
G#
G
Gb F #
F of concentration among musicians.
FIGURE 1. Response screens presented to the participant during the
test. Panel (a) shows the chroma response screen, with its 12 possible
responses equidistant from the center cursor position, and panel (b) shows
the octave response screen with each band of color (here shown in grey-
scale) corresponding to the C to B range of a single octave.
waveforms including a fundamental frequency and a
harmonic series (f, 2f, 3f, etc.) where the difference in
amplitude between adjacent harmonics was on the
order of 12 dB. Given a 16-bit sampling depth, this
yielded approximately 9 harmonics. All notes had a 1 s
duration including 50 ms linear onset and offset ramps.
Procedure
Participants were seated at a computer workstation and
given headphones through which to listen to the stimuli
at an intensity of approximately 75 dB SPL. The 108 trials
were presented in a randomized order and, upon hear-
ing each note, the participant saw the response screen
depicted in Figure 1a.1 After selecting a note name, the
participant was presented with a second response
screen (Figure 1b) from which s/he made an octave
judgment for the note identified with the previous
screen (color bands represent C to B of one octave).
Among general instructions given to the participant, it
was emphasized that this octave judgement does not
require the participant to select the exact note on the
grand staff but rather simply click anywhere in the col-
ored band representing the octave to which the note
belongs. Participants also were told that they should
respond as accurately and quickly as possible but were
left to use whichever strategy was most useful to them.
Trials were self-paced with no timeout and pauses were
introduced at the 1/3 and 2/3 marks through trials to
provide a short resting period. The timing of responses
was measured with an error inferior to 10 ms. Although
somewhat lacking in ethological validity, the chroma
response method chosen ensures that the mouse pointer
1
An earlier version of the response screen, used to collect the present
data, used only sharps rather than enharmonic notation.
resets to the center of the circle at the beginning of each
trial, thereby making each of the 12 possible responses
with the high temporal resolution, is critical for the col-
lection of reliable and unbiased reaction time data. Both
the stimulus type and response method also avoid con-
founds that might result from the different instruments
Scoring
Percent correct and mean absolute deviation (MAD)
were used as measures of accuracy. In the percent correct
score, only exactly correct responses are counted (0 semi-
tone deviation). MAD was computed by discarding the
octave information, taking the absolute value of the
semitone deviation from the correct answer (ranging
from -6 to +6 semitones) and then the mean of these
values across all trials for a given participant. Given a
possible range of 0 to 6 semitones deviation, a perfectly
random response pattern and therefore flat distribution
yields a MAD of 3, whereas a perfectly correct perform-
ance yields a MAD of 0. Reaction times are reported in
ms (after being iteratively cleaned by removing values
lying beyond 3 standard deviations of the mean) and as
log-transformed values (base 10).
Results
Participants
Data concerning age of commencement of training/
practice and total years of training/practice were not
available for 3 participants in each group. AP and NAP
groups did not differ in age (M = 23.33 years SE 0.71
vs. M = 23.14 years SE 0.64, respectively; t(49) = 0.85,
p = .40) or years of training/practice (M = 17.04 years
SE 0.79 vs. M = 16.09 years SE 0.80, respectively; t(43) =
0.84, p = 0.41). The AP group had a significantly lower
age of commencement of training/practice than the
NAP group (M = 5.46 years SE 0.35 vs. M = 6.95 years
SE 0.49, respectively; t(43) = 2.52, p = .02).
Split-Half Reliability
For each participant, the 108 randomly ordered trials
were split into two halves of 54 trials by selecting the odd
and even numbered trials separately. Correlations
between the two halves for the mean absolute deviation
and log-transformed reaction times (logRT) were r(49) =
.99, p < .001, and r(49) = .98, p < .001, respectively.
Music2702_02 11/16/09 6:00 PM Page 93

Absolute Pitch Ability 93

4 100

80

60

40

20

Mean Absolute Deviation (semitones)

3 0
6 5 4 3 2 1 0 1 2 3 4 5 6

100

80

60

2 40

20

100
0
6 5 4 3 2 1 0 1 2 3 4 5 6
80

60

40

1 20

0
6 5 4 3 2 1 0 1 2 3 4 5 6

0 20 40 60 80 100 120
Percent Correct
AP Subjects NAP Subjects

FIGURE 2. Percent correct versus mean absolute deviation for pitch class identification. Participants are grouped according to self-report and the
dotted-line box delimits the 95% confidence interval around chance performance. The inset histograms show examples of individual performance with
semitone deviation ranging 6 to +6 on the x-axis and number of responses on the y-axis.

Self-Report vs. Performance-Based Classification the accuracy, M = 78% SE 3.16 vs. M = 70% SE 3.35;
F(1, 49) = 2.83, p < .099, or log reaction times of octave
As can be seen in Figure 2, one participant who reported identification, M = 3.24 SE 0.03 vs. M = 3.23 SE
possessing AP performed more than 2 standard devia- 0.03; F(1, 49) = 0.14, p = .71. A repeated-measures
tions below the AP mean with a MAD of 1.94 semitones. ANOVA contrasting accuracy in identifying notes corre-
Conversely, another participant who self-reported as a sponding to white keys versus notes corresponding to
non-possessor achieved a MAD of 0.43 semitones, black keys and two levels of group (AP and NAP) yielded
more than 2 standard deviations above the NAP mean. a marginally significant interaction, F(1, 49) = 3.72, p =
These clear outliers therefore were reclassified according .06. Tukeys HSD posthoc tests revealed this interaction
to their performance. Results reported henceforth use to be due to significantly greater accuracy in identifying
this revised grouping. white notes among AP but not NAP participants
(Figure 4). The same was true for a similar contrast of
AP vs. NAP log reaction times for black and white notes, with a sig-
nificant interaction of group and note type, F(1, 49) =
For the identification of chroma, AP participants showed 23.12, p < .001, due to significantly faster reaction times
a greater number of correct responses, M = 77% SE 2.87 for white note identification among AP but not NAP
vs. M = 15% SE 3.04; F(1, 49) = 217.78, p < .001 participants (Figure 5). Additionally, accuracy and log
(Figures 3 and 4), smaller mean absolute deviation from reaction time performance were best for the pitch class
the correct chroma, M = 0.38 SE 0.10 vs. M = 2.48 A over all other notes in the AP group, F(1, 49) = 21.24,
SE 0.10; F(1, 49) = 221.24, p < .001, quicker reaction p < .001; F(1, 49) = 25.54, p < .001, respectively, as was
times, M = 3346 ms SE 526 ms vs. M = 7586 ms SE accuracy, F(1, 49) = 11.67, p = .001, but not log reaction
558 ms; F(1, 49) = 30.59, p < .001, and log reaction times, time in the NAP group. Age of commencement of train-
M = 3.45 SE 0.034 vs. M = 3.76 SE 0.04; F(1, 49) = ing was significantly correlated with MAD, r(43) = .46,
39.21, p < .001 (Figure 5) as compared to NAP partici- p = .01, percent correct chroma, r(43) = .44, p = .002,
pants. AP and NAP participants did not differ in either and log reaction time, r(42) = .40, p = .007. MAD also
Music2702_02 11/16/09 6:00 PM Page 94

94 Patrick Bermudez & Robert J. Zatorre

70

60

50
Percent of Total Responses

40

30

20

10

0
14 13 12 11 10 9 8 7 6 5 4 3 2 1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Semitone Deviation
NAP AP

FIGURE 3. Histogram of semitone deviations from the correct response. AP participants show the expected narrow modes around the exactly cor-
rect responses as well as the correct pitch class with an octave deviation (+ 12 and 12 semitones). The NAP musicians, on the whole, are not respond-
ing completely arbitrarily, with a very broad mode centered on correct identifications.

100

80

60
Percent Correct

40

20

C C# D D# E F F# G G# A A# B
Pitch Class
AP NAP

FIGURE 4. Percent correct identification for each pitch class by group (AP and NAP). The bands of grey correspond to the black keys of a piano or
non-diatonic notes of the C major scale. The pattern reveals how AP musicians identified diatonic notes more accurately on average.
Music2702_02 11/16/09 6:00 PM Page 95

Absolute Pitch Ability 95

4.0

Log Reaction Time 3.8

3.6

3.4

3.2
C C# D D# E F F# G G# A A# B
Pitch Class
AP NAP

FIGURE 5. The logRT for each pitch class by group (AP and NAP). The bands of grey correspond to the black keys of a piano or non-diatonic notes
of the C major scale. The pattern reveals how AP musicians identified diatonic notes more quickly on average.

was strongly correlated with log reaction time, r(49) = Discussion

.63, p < .0001 (Figure 6). Figure 7 plots percent correct
score against an index that multiplies mean absolute A number of issues associated with the behavioral
deviation by the log reaction times, thereby penalizing assessment of AP have obfuscated the distribution and
particularly time consuming strategies. variety of ability among both self-declared possessors

4
Mean Absolute Deviation (semitones)

2.8 3.0 3.2 3.4 3.6 3.8 4.0 4.2 4.4

Log Reaction Time
AP Subjects NAP Subjects

FIGURE 6. The logRT is plotted against mean absolute deviation time. Across all participants, the better performers clearly require less time to
respond. There is a trend towards longer reaction times among intermediate performers.
Music2702_02 11/16/09 6:00 PM Page 96
96 Patrick Bermudez & Robert J. Zatorre
55
Mean Absolute Deviation (semitones)
50
X Log Reaction Time
45
40
35
30
25
0 20
AP Subjects
FIGURE 7. Percent correct pitch class identification versus an index that multiplies mean absolute deviation by the logRT. This index effectively
weighs the MAD score with logRT so that more time consuming strategies are penalized. Despite this, a range of performance, including many inter-
mediate scores, is still obvious.
and non-possessors. The main design considerations
for our test of absolute pitch were intended to address a
few long-standing concerns in the community. For
instance, (1) we used multiharmonic synthetic stimuli
because they are equally unfamiliar to all listeners
(unlike specific instrument tones) and avoid problems
associated with the use of pure tones (loudness varia-
tion, poor performance for intermediate and weak AP
possessors, poor ethological validity), (2) we collected
both pitch class and octave judgements that allow us to
calculate the exact distance between the presented note
and the participants response and describe pitch-
height in addition to pitch-class judgements, (3) we
collected precise reaction times for both pitch class and
octave responses, allowing us to describe intermediate
performers who may be using alternate strategies to
those of the strongest performers, (4) we used a graph-
ical pitch class identification method that makes all 12
possible responses equally accessible and unbiased in
terms of timing and interface familiarity, thereby cir-
cumventing difficulties associated with keyboard-based
and other response methods, and (5) the wealth of col-
lected data allows us to derive indices of AP ability that
do not neglect performance variance that is beyond one
semitone from the correct response, as has often been
the case in the literature.
The data collected are indicative of the sensitivity of
the test, as it begins to make fine discriminations in
levels of absolute pitch proficiency. There are, of course,
40 60 80 100 120
Percent Correct
NAP Subjects
the striking differences in distributions of responses
between the AP and NAP groups (Figure 3) that dif-
fered in all the predicted ways. However, Figure 2 shows
how we begin to distinguish a third group of musicians
who typically do not consider themselves to have AP
but perform at a level intermediate to that of the clear
AP musicians (mean deviation < one semitone) and
NAP musicians with an essentially random response
(mean deviation ~ three semitones). These participants
also show a non-significant trend towards longer reac-
tion times (Figure 6) that would be consistent with the
use of more time consuming alternative strategies in
attempting to identify chroma. Greater numbers of
participants and trials in future work will be required to
tease apart those participants who are using a relative
pitch strategy from one or a few dependable notes versus
those who simply have broad, low-resolution chroma
representations centered on the correct pitch. The more
conventional measure of total correct chroma responses
does not discriminate as clearly between the intermediate
and weak performers (placing them closer to each other
than mean deviation; Figure 2), thereby highlighting
one of the limitations of this particular scoring strategy.
It is also notable that the split-half reliability of the test
is sufficiently strong so as to suggest that it is not only
internally reliable but also that a shortened version
would be suitably accurate for screening purposes. Two
participants, one claiming to possess AP and the other
not, performed in ways that were clearly discordant with
Music2702_02 11/16/09 6:00 PM Page 97

Absolute Pitch Ability 97

their self-report which highlights the need to verify AP including an important number of intermediate-level
performance explicitly although, on the whole, partici- performers (Figure 2). This is particularly clear if one
pants qualitative assessments of their abilities were considers a more comprehensive and intuitive descrip-
usually in reasonable agreement with test results. tor such as mean absolute deviation (MAD) from the
correct pitch class, a measure that ignores the octave
and therefore lies between 0 and 6 semitones and yields
Sampling and Distribution
an average of 3 for randomly distributed responses. The
result is a distribution of behavior across participants
Our recruitment effort advertised the need for musi-
that one would hesitate to describe as bimodal, though
cians both with and without absolute pitch. Although
one might construe mini-clusters among the very best
participants ultimately would report as either pos-
(deviation < 0.1) and worst (deviation ~ 3) performers.
sessing or not possessing AP in a forced-choice ques-
Most studies have awarded 1 point for exactly correct
tion (with certainty ratings), some non-possessors
pitch class identifications and a fraction of a point for
claimed that, under limited circumstances, they were
semitone deviations. Such approaches further exacerbate
able to label pitches with some degree of accuracy. A
the lack of distinction between those performing ran-
few of these and others who reported as either NAP
domly, with a MAD of approximately 3, and those with
or weak AP scored between 1.0 and 2.3 semitones
an average deviation of 2 semitones thus effectively
deviation, as can be seen in Figure 2. It is therefore
strengthening an apparent cluster of random performers.
interesting to note that, using a recruitment strategy
In our study, we cannot know for certain how participants
that welcomed a variety of ability, we have a consid-
responses might have differed had they been attempting
erable number of participants showing an intermedi-
to respond inside a typical 2 to 4 s time limit, though we
ate level of performance. There is some noticeable
can reasonably expect a higher error rate than is the case
clustering among the strongest (mean deviation ~ 0)
when the procedure is self-paced. Eight high-performing
and the weakest performers (mean deviation ~ 3) but
participants in our sample, scoring a mean absolute
also an important range of performance falling between
deviation of less than one, had mean reaction times rang-
these extremes. The trend remains clear even when
ing from 4 to 6 s and therefore many of their responses
considering the gross and strict measure of percent
would have earned a score of 0 under similar schemes
correct chroma identification. This is not entirely con-
(assuming that they were, in fact and as instructed,
sistent with the occasional claim of a clearly bimodal
performing as quickly as they could).
distribution among musicians (Athos et al., 2007;
Let us further concretize the effects of scoring with an
Carroll, 1975; Miyazaki, 1988), though there is wide-
example: two of our participants who performed at
spread acknowledgement that there exists a gamut of
chance according to percent correct chroma (each at
AP ability (Takeuchi & Hulse, 1993; Ward, 1999). The
8.3%) had very different mean absolute deviations of
appearance or not of a bimodal distribution seems to
1.44 and 2.81 semitones with mean reaction times of
depend in part on the way participants are recruited,
4905 and 9124 ms, respectively. The first of these par-
the test is administered, and the data are reduced,
ticipants more quickly gave a response that was, on
analyzed, and presented. Therefore, this remains a con-
average, much more accurate than the second partici-
troversial point and sources of discrepancy in results
pant. According to criteria used in many studies, these
may include sampling, testing materials, and scoring
participants would not be resolved (although they
strategies. Conjecture as to what differences in biology
would likely both be categorized as non-possessors).
may underlie performance in different regions of this
Assigning a partial or full point for semitone devia-
distribution is a very large topic unto itself (Baharloo
tions, as is often done in the AP literature to account for
et al., 2000; Bermudez, Lerch, Evans, & Zatorre, 2008;
semitone shifts/imprecision (see introduction), dimin-
Gregersen et al., 2000; Miyazaki, 2004; Ross et al., 2005;
ishes this effect but also can obfuscate important and
Zatorre, 2003).
subtle distinctions among high performing partici-
pants that are recovered by the mean absolute deviation
Scoring Strategy or other similar indices. (An index of overall performance
that combines accuracy and response latency information
Even when using the most stringent scoring strategy, might also be a useful descriptor; see y-axis of Figure 7.)
which only credits perfectly correct pitch class identifica- It can safely be suggested that a scoring scheme that
tion, a range of behavior from perfect to random perform- only considers perfectly correct responses (or, at most,
ance was observed across our sample of 51 musicians, semitone deviations) is likely to artificially accentuate
Music2702_02 11/16/09 6:00 PM Page 98
98 Patrick Bermudez & Robert J. Zatorre
differences between the top performers and others, and
sharpen the boundaries of a transition from strong to
weak performers by discarding important performance
variance. It follows from our results that if one allows
for a wider range of strategies in performing an AP task
and takes into consideration the full variability of accu-
racy, with less arbitrary descriptors as well as precise
reaction times, the apparent boundaries between differ-
ent groups of musicians may not be quite so manifest.
That said, one could construe from our data two small
clusters, one of 11 participants with nearly perfect per-
formance and another of 13 participants with essentially
random performance (though these are more apparent
on the percent correct axis than the MAD axis; Figure 2).
Ascertaining how these groups differ from each other,
and how each differs from intermediate performers, is
essential in the ongoing elucidation of the ability and
relating its characteristics to underlying neurological
infrastructure and genetics.
If we were to expect AP to manifest in a binary fash-
ion, we would observe a sharply bimodal distribution
of scores; no such effect is visible. If we were to expect
all intermediate performers to use a combination of AP
for a subset of notes and relative pitch calculation to fill
in the gaps, we should observe a bimodal distribution
for an index score that penalizes slower responses;
Figure 7 does not support this interpretation. If we use
a score that artificially penalizes anything but the
strongest performers, such as the strict percent correct
pitch class identification seen along the x-axis of Figure
2, we should accentuate any bimodal tendency; it is still
not clearly so. In sum, even though scoring strategies
that discard important proportions of performance
variance are likely to exaggerate and, in some cases, per-
haps even create the impression of bimodality in a dis-
tribution of responses, applying these harsh scoring
approaches to our data still does not create easily
resolvable subsets of participants. We can therefore say
with some confidence that there is true heterogeneity of
AP competence in our sample. Surely an important
contributor to this is the fact that we allowed our par-
ticipants an unlimited amount of time in which to give
their responses, allowing for the use of circuitous
strategies such as relying on absolute pitch for one or a
few notes and relative judgements to determine the rest.
Though this ability is often deprecated as pseudo-AP,
it is nonetheless of interest. In order for the relative
judgements to be accurate, the representation of the
absolute note(s) must be sufficiently accurate and of
high-resolution (narrow-band). All participants per-
forming with a mean deviation of less than one semi-
tone responded within 6 s, as did about half of the
random performers. Among NAP participants, there
was a non-significant trend towards longer response
times with lower MAD, perhaps indicating use of the
more time-consuming strategies among a subset of
intermediate performers. Figure 7, which plots percent
correct score against an index that multiplies mean
absolute deviation by the log reaction times (thereby
penalizing particularly time consuming strategies) still
yields a broad range of scores. This spectrum of ability
is behaviorally and (likely) biologically interesting.
Pitch Class Dependence
Several previously observed findings are replicated in
the data presented here. We are able to discern a pattern
where the notes corresponding to the C Ionian scale
(non-accidental or white piano keys) were identified
significantly more accurately and more quickly than
non-diatonic notes (accidental or black piano keys;
Miyazaki, 1988, 1989, 1990; Schellenberg, 2001;
Takeuchi & Hulse, 1991; Figures 4 and 5). Additionally,
a planned comparison revealed that the pitch class A,
the most typical tuning reference, was identified more
accurately and more quickly compared to the collapsed
performance across all other chroma. It was also more
accurately identified in the NAP group, likely driven by
the intermediate performers who may have used A as a
frequent reference for relative judgements of other
pitches. These differ from the findings of Miyazaki
(1990) showing that pitch classes C and G were most
quickly and accurately identified, presumably as a
reflection of their highest-ranking harmonic roles in a
C Ionian context. The cohort of musicians in the work
of Miyazaki (1990) was highly homogeneous in terms
of instrument of concentration (all keyboard players).
Our sample was broadly representative in that respect,
including keyboard, string, brass, percussion, and other
instrumentalists. Various instruments across these cat-
egories favor different keys in which notes occur with
different frequencies. Perhaps the only commonality
across them is the frequent use of A as a tuning refer-
ence. Also of note, the ratio of correct white to black
notes decreases as AP proficiency increases. In other
words, the very strongest performers are more likely to
label all chroma equally well (an obvious consequence
of making very few or no errors) whereas weaker per-
formers are more likely to identify white notes dispro-
portionately well. This is one of many lines of evidence,
along with the unusually high prevalence of AP among
the blind (Bachem, 1940; Hamilton, Pascual-Leone, &
Schlaug, 2004; Welch, 1988), East Asian music students
(Gregersen, Kowalsky, Kohn, & Marvin, 1999; Miyazaki,
Music2702_02 11/16/09 6:00 PM Page 99
1988, 2004), and musical savants (Heaton, 2003; Heaton,
Hermelin, & Pring, 1998; Lenhoff, Perales, & Hickok,
2001; Young & Nettelbeck, 1995), and broad perform-
ance distributions in our data and those of many other
researchers (Bachem, 1937; Takeuchi & Hulse, 1993;
Ward, 1999), contributing to the largely accepted notion
that most forms of AP are at least partly experience
dependent, with the typically more frequent exposure
to white notes during the early, formative years of music
education having a lasting impact on adult AP perform-
ance. This, of course, is in addition to the patently obvi-
ous and principled need for learning given that chroma
do not exist in nature; that is, they are not constituent
parts of the frequency continuum but rather impositions
of our scalar constructs. Therefore, in performing an
absolute pitch task such as ours, one must have formed
representations for the pitch classes to which the per-
ceived pitch is compared and subsequently labelled. As
has been suggested by Takeuchi and Hulse (1991), part
of the diatonic vs. non-diatonic effect may be due to
familiarity with the note names themselves, though no
work to date has disentangled the relative contributions
of pitch and note name/notation processing requirements
and familiarity.
AP and NAP participants did not differ in their
accuracy or speed of octave identification (Miyazaki,
1989). Though it might seem from Figure 3 as though
AP participants are committing more octave errors
given the peaks at 12 and +12 semitones deviation,
this a misleading impression resulting from the fact
that AP participants are able to make octave errors as
a direct consequence of being able to correctly iden-
tify chroma and thereby accumulate responses in the
same histogram bin. In contrast, the distribution of
responses among NAP musicians is largely unimodal
and a reflection of a coarser pitch height judgement.
AP participants are in fact non-significantly more
accurate in identifying octaves which we suspect is
due to the ability to accurately resolve notes close to
octave transitions (e.g., B3 vs. C4, with only a semi-
tone between them). Indeed, AP participants were
significantly more accurate in identifying the octave
assignments of B and C, t(49) = 2.22, p < .0001, but
not for the midway point in the octave F#, t(49) = 0.88,
p = .37. AP possessors have more recourse than tone
height in labelling octaves, being able to precisely locate
a note on a grand staff, yet, despite this, they do commit
octave errors that may in part be due to differences in
octave number and notation conventions between instru-
ments, lack of familiarity with the timbre of stimuli, and
overtone confusion (Takeuchi & Hulse, 1993; Ward &
Burns, 1982).
Absolute Pitch Ability 99
Conclusion
Although AP has been defined operationally within
some studies, typically by establishing an arbitrary per-
formance cut-off for an adhoc test, there are no com-
munity-wide conventions. That portion of disparity in
the literature that is not attributable to theoretical con-
siderations is counter-productive. A certain amount of
the discrepancy in the AP literature is doubtless due to
inconsistency in testing materials, methods, and crite-
ria, suggesting that there would be clear benefit from
standardization. If we hope to accurately describe the
AP ability, including possible subtypes and their rela-
tive incidences in musical populations, we should aim
to standardize our stimuli and administration parame-
ters. Doing so would allow us to test large numbers of
participants across many institutions and better
describe and ascertain the heterogeneity of AP. Also of
interest is how we would be afforded the opportunity to
compile norms that would then allow for statistically
meaningful descriptions and categorizations of per-
formance and, in turn, provide a means for opera-
tionally defining AP for a larger research community.
We stand to incur the usual advantages associated with
establishing a carefully designed standardized screening
and behavioral assessment battery, and convention
would permit comparisons between studies with a level
of confidence that has not been possible to date, there-
by hastening collective progress in the endeavor to
characterize and understand absolute pitch.
Ours is a traditional note-naming test that is (and
can only be) targeted to a population with considerable
knowledge, experience, and facility with conventional
Occidental note-naming schemes. This was a purpose-
ful design decision that suited our immediate purposes
as well as those of the vast majority of experimentalists
who specifically recruit highly trained musicians when
investigating AP. Relying on conventional musical
knowledge allows for the design of a simple, portable,
convenient, and discriminating test for AP ability
among trained musicians. The obvious disadvantage is
that it cannot identify any form of absolute pitch that
manifests itself outside the musical domain or inde-
pendently of musical instruction. Ross and colleagues
(Ross, Olson, Marks, & Gore, 2004) have described a
method appropriate for the identification of such indi-
viduals. A subset of their tasks generally take the form
of having a participant hear a tone (pure tone, iterated
ripple noise, or other) and, after a variable delay filled
with distractor tones, then use a variable frequency
tone generator to reproduce the presented tone. Ross
et al. (2005) have additionally posited a discrete biological
Music2702_02 11/16/09 6:00 PM Page 100

100 Patrick Bermudez & Robert J. Zatorre

mechanism that would account for the abilities of those
participants showing the strongest form of AP.
Our test has proved sensitive to key features of
absolute pitch proficiency. This is evinced by the detec-
tion of such effects as greater accuracy and speed of
pitch class identification for white versus black notes
(C major diatonic vs. non-diatonic notes), most accu-
rate and fastest identification for the pitch class A, and
the description of a wide variety of absolute pitch pro-
ficiency. Having provided a validation of our AP test,
we are now proceeding to the next step in development:
we intend to produce a distributable version for the PC
that will be available to other centers and researchers to
provide a reliable and precise method for behaviorally
assessing absolute pitch ability in musicians. The test will
be distributed through the BRAMS Laboratory (Montral)
upon request. It is our hope that this will constitute a
modest step toward the above mentioned benefits and
avoid experimenters unnecessary development costs for
proprietary tests. Overall, the aim is to gather multidi-
mensional information (accuracy, latency, per-note
patterns of response, etc.) across many research centers
in order to more comprehensively describe absolute pitch
and provide the means for identifying possible subpop-
ulations, as well as deriving behavioral indices that can
be regressed onto various biological metrics. This may
help in deciphering the behavioral characteristics and
neural underpinnings of this unusual ability which serves
as an interesting model for several aspects of cerebral
function.
Author Note
We thank Marc Bouffard, Pierre Ahad, and Joyce Chen
for their contributions and consultations. This work
was funded by the International Foundation for Music
Research (IFMR), the Canadian Institutes of Health
Research (CIHR), and the Natural Sciences and
Engineering Research Council of Canada
Correspondence concerning this article should be
addressed to Patrick Bermudez, Department of Cognitive
Neuroscience, Montreal Neurological Institute, 3801
University, room 276, Montral, Qubec, Canada H3A
2B4. E-MAIL: patrick@bic.mni.mcgill.ca

References

ATHOS , E. A., L EVINSON , B., K ISTLER , A., Z EMANSKY, J.,
B OSTROM , A., F REIMER , N., ET AL . (2007). Dichotomy and
perceptual distortions in absolute pitch ability. Proceedings of
the National Academy of Sciences, USA, 104, 14795-14800.
B ACHEM , A. (1937). Various types of absolute pitch. Journal of
the Acoustical Society of America, 9, 146-151.
B ACHEM , A. (1940). The genesis of absolute pitch. Journal of the
Acoustical Society of America, 11, 434-439.
BAHARLOO, S., JOHNSTON, P. A., SERVICE, S. K., GITSCHIER, J., &
FREIMER, N. B. (1998). Absolute pitch: An approach for identi-
fication of genetic and nongenetic components. American
Journal of Human Genetics, 62, 224-231.
BAHARLOO, S., SERVICE, S. K., RISCH, N., GITSCHIER, J., & FREIMER,
N. B. (2000). Familial aggregation of absolute pitch. American
Journal of Human Genetics, 67, 755-758.
B ENJAMIN , P. A. (2008). Transmission and population genetics,
(2nd ed.). Basingstoke, Hampshire: W. H. Freeman.
B ERMUDEZ , P. (2008). The neural correlates of absolute pitch.
Unpublished doctoral dissertation, McGill University, Montreal.
B ERMUDEZ , P., L ERCH , J. P., E VANS , A. C., & Z ATORRE , R. J.
(2008). Neuroanatomical correlates of musicianship as
revealed by cortical thickness and voxel-based morphometry.
Cerebral Cortex, 19, 1583-1596.
B ERMUDEZ , P., & Z ATORRE , R. J. (2009). The absolute pitch
mind continues to reveal itself. Journal of Biology, 8, 75.
C ARROLL , J. B. (1975). Speed and accuracy of absolute pitch
judgments: Some latter day results. Educational Testing
Service Research Bulletin, 6, 1-77.
C HIN , C. S. (2003). The development of absolute pitch: A
theory concerning the roles of music training at an early
developmental age and individual cognitive style. Psychology
of Music, 31, 155-171.
C OSTA-G IOMI , E., G ILMOUR , R., S IDDELL , J., & L EFEBVRE ,
E. (2001). Absolute pitch, early musical instruction, and
spatial abilities. Annals of the New York Academy of Sciences,
930, 394-396.
G REGERSEN , P., KOWALSKY, E., KOHN , N., & M ARVIN , E.
(2000). Early childhood music education and predisposition
to absolute pitch: teasing apart genes and environment.
American Journal of Human Genetics A, 98, 280-282.
GREGERSEN, P. K., KOWALSKY, E., KOHN , N., & MARVIN , E. W.
(1999). Absolute pitch: Prevalence, ethnic variation, and
estimation of the genetic component. American Journal of
Human Genetics, 65, 911-913.
H AMILTON , R. H., PASCUAL -L EONE , A., & S CHLAUG , G. (2004).
Absolute pitch in blind musicians. Neuroreport, 15, 803-
806.
H EATON , P. (2003). Pitch memory, labelling and disembedding
in autism. The Journal of Child Psychology and Psychiatry, 44,
543-551.
Music2702_02 11/16/09 6:00 PM Page 101
H EATON , P., H ERMELIN , B., & P RING , L. (1998). Autism and
pitch processing: A precursor for savant musical ability. Music
Perception, 15, 291-305.
K APLAN , R. M., & S ACCUZZO, D. P. (1993). Psychological
testing: Principles, applications, and issues (3rd ed.). Pacific
Grove, California: Books/Cole.
L ENHOFF, H. M., P ERALES , O., & H ICKOK , G. (2001). Absolute
pitch in Williams syndrome. Music Perception, 18, 491-503.
L EVITIN , D. J., & R OGERS , S. E. (2005). Absolute pitch:
Perception, coding, and controversies. Trends in Cognitive
Sciences, 9, 26-33.
M ILLER , G. A. (1956). The magical number 7, plus or minus
2Some limits on our capacity for processing information.
Psychological Review, 63, 81-97.
M IYAZAKI , K. (1988). Musical pitch identification by absolute
pitch possessors. Perception and Psychophysics, 44, 501-512.
M IYAZAKI , K. (1989). Absolute pitch identificationEffects of
timber and pitch region. Music Perception, 7, 1-14.
M IYAZAKI , K. (1990). The speed of musical pitch identification
by absolute-pitch possessors. Music Perception, 8, 177-188.
M IYAZAKI , K. (2004). How well do we understand absolute
pitch? Acoustical Science and Technology, 25, 426-432.
M ORENO S ALA , M. T., & C OSTA-G IOMI , E. (2008). Cognitive
and spatial abilities related to young childrens development of
absolute pitch perception.Unpublished manuscript.
NLM. (2008). Genes and disease. Retrieved June 2008 from http://
www.ncbi.nlm.nih.gov/books/bv.fcgi?rid=gnd
P OLLACK , I. (1952). The information of elementary auditory
displays. Journal of the Acoustical Society of America, 24,
745-749.
Absolute Pitch Ability 101
R OSS , D. A., G ORE , J. C., & M ARKS , L. E. (2005). Absolute
pitch: Music and beyond. Epilepsy and Behavior, 7, 578-601.
ROSS , D. A., O LSON , I. R., M ARKS , L. E., & G ORE , J. C. (2004).
A nonmusical paradigm for identifying absolute pitch possess-
ors. Journal of the Acoustical Society of America, 116, 1793-1799.
S CHELLENBERG , E. G. (2001). Asymmetries in the discrimination
of musical intervals: Going out-of-tune is more noticeable than
going in-tune. Music Perception, 19, 223-248.
TAKEUCHI , A. H., & H ULSE , S. H. (1991). Absolute-pitch judg-
ments of black-key and white-key pitches. Music Perception, 9,
27-46.
TAKEUCHI , A. H., & H ULSE , S. H. (1993). Absolute pitch.
Psychological Bulletin, 113, 345-361.
VAN K REVELEN , A. (1951). The ability to make absolute judg-
ments of pitch. Journal of Experimental Psychology, 42, 207-215.
WARD, W. D. (1999). Absolute pitch. In D. Deutsch (Ed.), The
psychology of music (2nd ed., pp. 265-298). San Diego:
Academic Press.
WARD, W. D., & B URNS , E. M. (1982). Absolute pitch. In D.
Deutsch (Ed.), The psychology of music (1st ed., pp. 431-451).
San Diego: Academic Press.
W ELCH , G. F. (1988). Observations on the incidence of
absolute pitch (AP) ability in the early blind. Psychology of
Music, 16, 77-80.
YOUNG , R. L., & N ETTELBECK , T. (1995). The abilities of a
musical savant and his family. Journal of Autism and Develop-
mental Disorders, 25, 231-248.
ZATORRE, R. J. (2003). Absolute pitch: A model for understanding
the influence of genes and development on neural and
cognitive function. Nature Neuroscience, 6, 692-695.
Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.