ARTICLE IN PRESS

Sensorimotor Mismapping in Poor-pitch Singing

Hao He and Wei-dong Zhang, Shanghai, China

Summary: Objective. This study proposes that there are two types of sensorimotor mismapping in poor-pitch singing:
erroneous mapping and no mapping. We created operational definitions for the two types of mismapping based on the
precision of pitch-matching and predicted that in the two types of mismapping, phonation differs in terms of accuracy
and the dependence on the articulation consistency between the target and the intended vocal action. The study aimed
to test this hypothesis by examining the reliability and criterion-related validity of the operational definitions.
Study Design. A within-subject design was used in this study.
Methods. Thirty-two participants identified as poor-pitch singers were instructed to vocally imitate pure tones and
to imitate their own vocal recordings with the same articulation as self-targets and with different articulation from self-targets.
Results. Definitions of the types of mismapping were demonstrated to be reliable with the split-half approach and to
have good criterion-related validity with findings that pitch-matching with no mapping was less accurate and more de-
pendent on the articulation consistency between the target and the intended vocal action than pitch-matching with erroneous
mapping was. Furthermore, the precision of pitch-matching was positively associated with its accuracy and its depen-
dence on articulation consistency when mismapping was analyzed on a continuum. Additionally, the data indicated
that the self-imitation advantage was a function of articulation consistency.
Conclusion. Types of sensorimotor mismapping lead to pitch-matching that differs in accuracy and its dependence
on the articulation consistency between the target and the intended vocal action. Additionally, articulation consistency
produces the self-advantage.
Key Words: Poor-pitch singing–Pitch-matching–Sensorimotor mapping–Precision–Articulation.

INTRODUCTION The cutoff that defines an acceptable range of precision scores

Individuals who are unable to match musical pitch accurately
or precisely are considered to have poor-pitch singing.1,2 Re-
searchers measure the proficiency of pitch singing by using pitch-
matching tasks in which participants are instructed to vocally
imitate target pitches.3–6 In terms of objective measurement based
on acoustic methods that extract the fundamental frequency (F0)
of a sound, two measures are used to evaluate the pitch-
matching performance. One is accuracy, which refers to the
average difference between the target pitch and the pitch that
an individual sings. Accuracy scores represent the proximity of
vocally produced pitches to the target pitches. Accuracy is mea-
sured in cents or semitones.a Because the human song system
is highly complicated and involves peripheral mechanisms of vo-
calization and relevant neural networks,7,8 it is less likely for an
individual to exactly match a target pitch. Thus, an acceptable
range of pitch deviation must be determined for accurate pitch-
matching. The cutoff for acceptable deviation of a produced pitch
from the target pitch ranges from 100 cents2,5 to 50 cents.3,4,6
Another measure for pitch-matching is precision. Precision, which
in statistics refers to the standard deviation (SD) of sung pitches,
measures the variability of pitch-matching across repeated trials.
Accepted for publication February 24, 2017.
Conflict of interest: The authors have no conflict of interest to declare.
a
Both semitone and cent are the units of interval (the difference between two pitches)
used in the modern Western tuning system. In mathematics, a semitone is derived from
dividing an octave (with a frequency ratio of 2:1) into 12 equal parts, and a cent is derived
from dividing a semitone into 100 equal parts.
From the School of Psychology and Cognitive Science, East China Normal University,
Shanghai, China.
Address correspondence and reprint requests to Wei-dong Zhang, School of Psychology
and Cognitive Science, East China Normal University, No. 3663. North Zhongshan Road,
Shanghai 200062, China. E-mail: wdzhang@psy.ecnu.edu.cn
Journal of Voice, Vol. ■■, No. ■■, pp. ■■-■■
0892-1997
© 2017 The Voice Foundation. Published by Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.jvoice.2017.02.018
for precise pitch-matching is set at 100 cents or 50 cents.2,3
Mechanism of poor-pitch singing: background
According to previous models of the human song system,7,9–11
poor-pitch singing can be caused by a dysfunction of
perception, 12–18 sensorimotor integration, 4–6,19 vocal motor
control,4,20,21 or memory.15,22 Among these, a malfunction in sen-
sorimotor integration is thought to be the main cause of poor-
pitch singing.4,5
Regarding the mechanism of sensorimotor dysfunction in poor-
pitch singing, an early explanation suggested that sensorimotor
mismapping between pitch percepts and phonatory gestures
caused systematic singing errors. The explanation accounted for
the phenomenon of interval compression that occurred while
singing melodies,5,23 but it was challenged based on the consid-
erable variability in pitch-matching.2–5 Given this limitation,
researchers introduced the internal model framework based on
the domains of motor planning and control,24 and proposed the
inverse modeling mechanism, arguing that poor-pitch singers lack
sensorimotor translation between pitch percepts and phonatory
gestures.6 However, the inverse modeling account was not more
informative than the previous sensorimotor mismapping account
in terms of explaining how the sensorimotor mismapping forms.
A recent multimodal imagery association (MMIA) model
offered some answers to this question.19 In the MMIA model,
comfort pitch, the pitch that an individual sings comfortably, plays
a role in distorting the initially unbiased sensorimotor mapping
via its attracting influence on pitch-matching. Mathematically,
the distortion is explained as the convolution of two probabil-
ity distributions: (1) an unbiased sung pitch distribution that
centers on the target pitch and (2) a comfort pitch distribution.
The MMIA model explains not only systematic bias but also the
 ARTICLE IN PRESS
2 Journal of Voice, Vol. ■■, No. ■■, 2017
large variability in pitch-matching that was observed in previ-
ous studies.3–5
The MMIA model, in a mathematical way, sheds light on the
formation mechanism of sensorimotor mismapping in poor-
pitch singing. However, little research has elaborated the construct
of the sensorimotor mapping system. In fact, the four catego-
ries of pitch-singing derived from the combination of accuracy
(accurate vs. inaccurate) and precision (precise vs. imprecise),
which have been systematically examined in previous studies,2,3
suggest that there may be multiple types of sensorimotor mapping.
In particular, because pitch-matching can be either imprecise (both
inaccurate and accurate) or inaccurate yet precise, there are likely
different types of mismapping.
A new hypothesis
With the aim of exploring the construct of mismapping at the
category level, this study proposes the hypothesis that senso-
rimotor translation involves a mapping system that encompasses
three types of mapping: accurate mapping, erroneous mapping,
and no mapping. Accurate mapping typically leads to accurate
and precise pitch-matching. With erroneous mapping, an indi-
vidual forms a fixed but inaccurate connection between a pitch
percept and a phonatory gesture that does not produce the pitch.
When erroneous mapping occurs, the sung pitch varies little from
a certain pitch height (but is not the same as the target), pro-
ducing inaccurate but precise pitch-matching. With no mapping,
there is no formed connection between a pitch percept and a pho-
natory gesture. In such cases, an individual may randomly map
a pitch percept onto a phonatory gesture each time he or she imi-
tates the target, producing imprecise pitch-matching across
repeated trials.
This study then proposes an operational definition of each type
of mapping based on the relationship between accuracy and pre-
cision. Accuracy is measured as the absolute value of the mean
signed pitch deviation, whereas precision is measured as the SD
of signed pitch deviation. Because a specific cutoff value may
be more sensitive to precision than accuracy,2,3 we established
a 50-cent cutoff for accuracy and a more liberal cutoff of 100
cents for precision in this study. Using these cutoff values, ac-
curate and precise pitch-matching represents correct auditory-
motor mapping; inaccurate but precise pitch-matching represents
erroneous mapping; and imprecise pitch-matching, regardless of
whether it is accurate, represents no mapping. It is worth noting
that an individual can exhibit both types of sensorimotor
mismapping simultaneously or just one of them.
Hypothesis testing
The proposed sensorimotor mismapping hypothesis can be tested
by examining the reliability and validity of the operational defi-
nitions of the types of mismapping. For reliability testing, this
study used the split-half approach, which tests the degree of cor-
relation between data of the first and the second halves.25 In
particular, we analyzed reliability by splitting trials into two equal
parts in terms of time order and investigating the correlation
between the parts.
To assess validity, we used criterion-related validity, which
assumes that there is a correlation between a test measure and
one or more external criteria.26 Two criteria were applied in this
study. The first criterion was the accuracy of pitch-matching. Pitch-
matching with erroneous mapping should be more accurate than
pitch-matching with no mapping for two possible reasons. First,
erroneous mapping may be the outcome of a compromise after
multiple failed attempts to match a target pitch. Although indi-
viduals may know that they make singing errors, they may be
unable at the time to improve their singing accuracy; thus, they
must accept their most accurate vocal action for the sake of ex-
pediency. Considering that the majority of poor-pitch singers have
normal pitch perception,4,5 the produced pitch deviation would
not be considerable in such cases. Second, erroneous mapping
can result from limited perceptual resolution of human vocal-
izations. Previous research has shown that the differential threshold
is higher when the targets are natural human voices than when
the targets are synthesized vocal sounds.4 Thus, for individuals
who have normal pitch perception, the failure to detect devia-
tions in the produced pitch through auditory feedback can lead
to erroneous mapping, but their accuracy would not be substan-
tially affected. In contrast, no mapping can cause large amplitudes
of pitch deviations, as in cases when individuals attempt to match
pitches by trial and error and still do not succeed.
The second criterion for the validity test was the depen-
dence of phonation on the articulation consistency between the
target and the intended vocal action. Our starting point was a
widely recognized finding that phonation and articulation are
structurally linked. Early research has shown that a change in
articulation changes the produced pitch.27–29 Additionally, the ar-
ticulation consistency between the target and the vocal action
can affect the accuracy of pitch-matching. As research has shown,
pitch-matching is more accurate when human voices are imi-
tated than when tones produced by instruments are imitated,4,30–32
and it is the most accurate in self-imitation.4,6,31 This is the so-
called human voice or self-imitation advantage. Note also that
the timbre similarity explanation has been ruled out by previ-
ous research.6 The increased accuracy of pitch-matching when
imitating human voices and during self-imitation suggests that
phonation is affected by the articulation consistency between the
target and the intended vocal action. In other words, the human
voice or self-imitation advantage may be a function of articu-
lation consistency.
In the sensorimotor mismapping hypothesis proposed in this
study, phonation is more dependent on the articulation consis-
tency in no mapping than in erroneous mapping for the following
reasons. In pitch imitation, a phonatory solution is needed when
an auditory input is received. In erroneous mapping, a phona-
tory motor plan can be quickly generated because of the
sensorimotor mappings that are formed. In no mapping, the plan-
ning of phonation is not directly driven by a formed sensorimotor
association, but it can be affected by external vocal motor cues.
Articulation, a set of complicated and rapid motor behaviors in-
volving the coordination of laryngeal, pharyngeal, and orofacial
muscles, can provide such cues. Articulation is more imitable
than phonation because articulation, which comprises such com-
ponents as lip, tongue, and jaw movements, can be observed,
whereas phonation lies deep in the throat and is difficult to
observe. Thus, the unplanned phonation in no mapping is more
 ARTICLE IN PRESS
Hao He and Wei-dong Zhang Sensorimotor Mismapping in Pitch-matching
FIGURE 1. Flowchart illustrating the rationale for the validity test using the criterion of articulation consistency. Boxes indicate the three pitch-
matching tasks: the pure tone imitation (PTI) task, the same articulatory self-imitation (SASI) task, and the different articulatory self-imitation
(DASI) task. Ovals indicate the self-advantage measures attributed to changes in the articulation consistency between the target and the intended
vocal action. Dashed braces indicate the comparisons between measures.
dependent on the articulation consistency between the target and
the intended vocal actions than the planned phonation in erro-
neous mapping is.
Operationally, the dependence of phonation on the articula-
tion consistency between the target and the intended vocal actions
was estimated with two measures in this study. The first measure
was the size of the self-imitation advantage. According to the
proposed sensorimotor mismapping hypothesis, the self-
advantage should be greater in no mapping than in erroneous
mapping. The inclusion of the self-advantage was based on the
above-proposed hypothesis that the articulation consistency
between the target and the intended vocal action produces the
self-advantage. In this study, this hypothesis was tested before
the validity test that used the criterion of phonation’s depen-
dence on articulation consistency was applied. The second
measure for the dependence of phonation on articulation con-
sistency was the linear relationship between the self-advantage
and the precision of pitch-matching, which included the mod-
erating effect of sensorimotor mapping. The difference in this
linear relationship between the two types of mismapping may
take one of the following two forms: First, the linear relation-
ship may exist in no mapping but not in erroneous mapping.
Second, if the linear relationship exists in both types of
mismapping, the slope of the regression line should be larger
with no mapping than with erroneous mapping.
The current experiment
In the present study, we designed a pure tone imitation (PTI)
task, a same articulatory self-imitation (SASI) task, and a dif-
ferent articulatory self-imitation (DASI) task. The participants
were instructed to imitate pure tones in the PTI task and to imitate
their own vocal recordings with the same articulation as self-
targets in the SASI task and with different articulation from self-
targets in the DASI task. Assuming that mismapping (precision)
is a state variable that is sensitive to task, we tested the relia-
bility and the criterion-related validity of using pitch-matching
3
accuracy as a criterion for the operational definition of
mismapping types in each of the three tasks. For the validity test
using the dependence of phonation on the articulation consis-
tency between the target and the intended vocal action as a
criterion, Figure 1 illustrates the course and its rationale. The
PTI task served as a baseline test. Two types of self-advantage
were produced by comparing the participants’ performance on
the PTI task with their performance on the SASI and the DASI.
The difference between the two types of self-advantage repre-
sented the variations in pitch-matching (based on the PTI task)
between the SASI task and the DASI task, which was caused
by the changes in articulation consistency between the target and
the intended vocal action and reflected the dependence of pho-
nation on the articulation consistency. Additionally, we included
pitch-matching with accurate mapping as a control condition.
According to the proposed mismapping hypothesis, pitch-
matching with accurate mapping and pitch-matching with
erroneous mapping should not differ in their dependence on ar-
ticulation consistency, and pitch-matching with the two types
of mapping should be less dependent on articulation consisten-
cy than pitch-matching with no mapping.
METHODS
Participants
The participants were recruited through an advertisement posted
on campus for students who thought they were poor singers or
wished to know whether they were poor singers. Thirty-two par-
ticipants were identified as poor-pitch singers after a series of
tests (detailed descriptions are presented in the procedure section).
The participants (26 women and 6 men aged between 18 and
31 years old; M = 23.3, SD = 3.1) had never received formal
musical training, except two who reported 4 and 14 years of in-
strument training. None of the participants reported having been
diagnosed with hearing or vocal motor deficits, brain lesions or
mental disorders, or having a history of surgery.
 ARTICLE IN PRESS
4 Journal of Voice, Vol. ■■, No. ■■, 2017
Materials
In the PTI task, for each participant, the target tones were 12
component pitches on a chromatic scale within the partici-
pant’s vocal range. All the targets were sine waves with a 1-second
duration. In principle, each participant’s target chromatic scale
was centered within the participant’s vocal range (between the
lowest and the highest naturally produced pitches). However, to
guarantee the homogeneity of the stimuli for most partici-
pants, a chromatic scale from C4 (261.63 Hz) to B4 (493.88 Hz)
was selected for female participants who had a vocal range equal
to or wider than the C4–B4 range, and a scale ranging from C3
(130.81 Hz) to B3 (246.94 Hz) was selected for male partici-
pants with a vocal range equal to or wider than the C3–B3 range.
These two chromatic scales were used for three-quarters of the
participants.
In the SASI task and the DASI task, the targets for each par-
ticipant were his or her own vocal recordings from the PTI task.
Operationally, for each participant and at each scale degree, we
selected a recording with a stably produced pitch, modulated it
to the standard pitch, and then ensured that all targets were pre-
sented at a similar volume.
Apparatus
The pure tones were generated with Adobe Audition 1.5 (Adobe,
San Jose, CA) with a sampling rate of 44,100 Hz and a 16-bit
resolution and were presented through AKG K99 headphones
(AKG Acoustics GMBH, Vienna, Austria). The tones that the
participants produced were recorded using an ISK BM-800 con-
denser microphone (Ningbo Shengke Electronic Co., Ltd., Ningbo,
China).
Procedures and tasks
The experiment included two phases. The first phase com-
prised the screening tests (including two pitch perception tasks
and a vocal sweep task) and the PTI task. The second phase con-
sisted of two self-imitation tasks (SASI and DASI) and was
conducted 3 days after the first phase. All the tasks were carried
out in a quiet room. The first phase lasted approximately 50
minutes, and the second phase lasted approximately 40 minutes.
Pitch perception tasks
This session aimed to measure the participants’ pitch discrim-
ination thresholds using the maximum likelihood procedure.33
In each trial, the participants heard two pure tones sequen-
tially. All the tones lasted 1 second, and the two tones presented
in each trial were separated by 250 ms. Each subject com-
pleted two tasks, each of which comprised three blocks of 25
trials. In the first task, the participants were asked to decide
whether the two tones were the same. In this task, 20% of the
trials were set as catch trials, in which the two tones had the same
pitch. In the second task, the participants were required to in-
dicate which of the two tones was higher. For all the participants
and both tasks, C4 (261.63 Hz) was set as the standard frequen-
cy. The thresholds were calculated at the end of each block using
the maximum likelihood procedure, and individuals with average
thresholds above 50 cents for either of the two pitch percep-
tion tasks were excluded.
Vocal sweep task
This task aimed to identify whether the participants had a vocal
motor deficit. It also served as a vocal warm-up. The partici-
pants were instructed to sing naturally from the lowest pitch to
the highest pitch they could produce in small steps and then to
reverse this process. The participants performed this procedure
five times. Vocal motor proficiency was estimated by measur-
ing whether a participant could produce all of the component
pitches on a chromatic scale within his or her vocal range. As
the criterion for producing a pitch, we required a sung tone to
maintain a stable pitch height and to last for more than 400 ms.
The F0s of the sung tones were quickly measured using Adobe
Audition (Adobe).
Pure tone imitation task
In the PTI task, each trial started with a pure tone (A6, 1760 Hz)
cue that lasted 250 ms. After 750 ms of silence, the target tone
was presented for 1 second. The intertrial interval was 4 seconds.
During this temporal blank period, the participants imitated target
pitches with the syllable /ba/ as accurately and quickly as pos-
sible. They were asked (1) to keep singing using a stable pitch
and volume, (2) to correct their singing when they perceived that
they were incorrect, and (3) to keep singing for a bit longer than
the target. The sung tones were saved as .wav files. The proce-
dure comprised 12 blocks. Each block consisted of 12 trials in
which pitches were presented for all degrees of a chromatic scale
(ie, I, II, III. . .XII). In each block, target pitches at 12 degrees
of a chromatic scale were pseudorandomly presented, and the
blocks were also presented in pseudorandom order.
Same or different articulatory self-imitation task
In the SASI task, after hearing the target sung with the syllable
/ba/, the participants were asked to imitate the target pitch with
/ba/ as accurately and quickly as possible. In contrast, in the DASI
task, the participants were required to imitate the target pitch
with the syllable /di/ as accurately and quickly as possible. Half
of the participants performed the SASI task first and then per-
formed the DASI task, and the other half performed the tasks
in reverse order. For each participant, the stimuli were pre-
sented in the same manner and order as in the PTI task.
Data analysis
We conducted an acoustic analysis of the vocal recordings using
the MIR toolbox34 in MATLAB (MathWorks, Natick, MA), with
a sampling rate of 44,100 Hz and a frame decomposition rate
of 100 Hz. For this analysis, we measured the F0s of the vocal
recordings and converted the unit of pitch from hertz to cents
to perform meaningful comparisons in the 12-tone equal
temperament.
We computed the absolute value of the mean signed pitch de-
viation to estimate the accuracy of pitch-matching and computed
the SD of signed pitch deviations to estimate the precision of
pitch-matching. In this study, the participants either with PTI
accuracy scores greater than 50 cents or with PTI precision scores
greater than 100 cents were identified as poor-pitch singers. As
previously described, the participants were allowed to adjust their
pitch repeatedly within the 4-second response time window.
 ARTICLE IN PRESS
Hao He and Wei-dong Zhang Sensorimotor Mismapping in Pitch-matching
Hence, in each trial, we measured the F0 of the last sung tone.
In fact, the participants in this study seldom imitated the target
more than once.
After data preprocessing, each participant had one data point
in the subject dimension and 12 data points (corresponding to
the 12 target scale degrees of a chromatic scale) in the pitch height
dimension for each observed variable (accuracy and precision).
We used the subject-level data for analyzing the self-advantage.
We used the pitch height-level data for the reliability and va-
lidity tests of the operational definitions of erroneous mapping
and no mapping. The pitch height-level data were analyzed using
mixed linear models because their hierarchical structures violate
the assumptions of independence for analysis of variance (ANOVA)
and multiple regression.35 We conducted the statistical analysis
using the SPSS Statistics version 22 (IBM Corp., Armonk, NY).
RESULTS
General performance of the participants
The 32 participants were defined as poor-pitch singers accord-
ing to their performance on the PTI task (the baseline condition).
The average accuracy score for the PTI was 184 cents (SD = 116
cents), and the average precision score was 129 cents (SD = 59
cents). Twenty-one of the participants demonstrated inaccurate
and imprecise pitch-matching, and the rest demonstrated inac-
curate yet precise pitch-matching. All the participants showed
normal pitch perception. The average threshold for the same or
different judgments was 30 cents (SD = 10 cents). The average
threshold for the higher tone identifications was 24 cents (SD = 12
cents). None of the participants showed vocal motor problems,
as they all passed the vocal sweep task. According to models
of the processes of pitch singing,7,9–11 the present data sug-
gested that the participants’ poor pitch singing was the result
of a dysfunction of sensorimotor association rather than a dis-
order of pitch perception or vocal motor control or memory.
Reliability test
For each pitch-matching task, we divided the 12 blocks equally
into two sections comprising the first six blocks and the last six
blocks. Because all the component pitches on a chromatic scale
were presented once in each block, both sections had an iden-
tical number of each target pitch. For each participant, we
computed the precision score of the pitch-matching at each target
pitch-height in the first and second sections. Afterward, we ana-
lyzed the overall correlation between the two sections with respect
to precision. The results showed that the Pearson correlation co-
efficient was 0.588 (P < 0.001) for the PTI, 0.675 (P < 0.001)
for the SASI, and 0.725 (P < 0.001) for the DASI. Similar results
were obtained for the single-measure intraclass correlation co-
efficient (ICC) for absolute agreement (PTI: ICC = 0.534,
P < 0.001; SASI: ICC = 0.674, P < 0.001; DASI: ICC = 0.724,
P < 0.001). Given that the datasets had hierarchical structures,
we then constructed mixed linear models to explore whether there
was a linear relationship between the two sections with respect
to precision. In the ultimate modeling solution for each dataset,
the model used the precision scores of the second section as the
dependent variable, the scale degree (I–XII) as the fixed factor,
5
the precision scores of the first section as the covariate, and the
fixed and random intercepts for participants. The results showed
a significant linear relationship between the precision of the first
section and that of the second section for each task (PTI:
b = 0.277, SE = 0.043, P < 0.001; SASI: b = 0.375, SE = 0.045,
P < 0.001; DASI: b = 0.363, SE = 0.052, P < 0.001). The data
suggested that the operational definitions of the types of
mismapping have moderate reliability.
Validity test
Criterion 1: accuracy
For each task, we constructed a 12 (scale degree: I–XII) × 2 (type
of mismapping: erroneous mapping vs. no mapping) mixed linear
model. In the model, we set accurate scores of pitch-matching
as the dependent variable and the scale degree and type of
mismapping as the fixed factors, and we set the fixed and random
intercepts for participants. For the PTI, a significant main effect
of type of mismapping was revealed (F[3,319] = 27.401,
P < 0.001), in which erroneous mapping was associated with more
accurate pitch-matching than no mapping. However, there was
no significant main effect of type of mismapping for either the
SASI or the DASI (Fs < 3.668, Ps > 0.050).
Self-imitation advantage
Before conducting the validity test, in which we used the self-
imitation advantage to measure the dependence of phonation on
the articulation consistency under the two types of mismapping,
we needed to test the proposed hypothesis that articulation con-
sistency produces the self-advantage. To do so, we performed
two repeated-measures ANOVAs for accuracy and precision of
pitch-matching, setting the task (PTI, SASI, and DASI) as the
within-subjects factor. All the reported values were Greenhouse-
Geisser corrected. Both ANOVAs showed significant main effects
of task (ie, the self-advantage) on accuracy (F[1,33] = 61.800,
P < 0.001, ηp2 = 0.667) and precision (F[2,47] = 77.374, P < 0.001,
ηp2 = 0.714). Multiple comparisons using the Bonferroni cor-
rection revealed significant differences in the two measures
between each two pitch-matching tasks. As Figure 2 shows, the
FIGURE 2. The accuracy scores and the precision scores of pitch-
matching in the pure tone imitation (PTI) task, the same articulatory
self-imitation (SASI) task, and the different articulatory self-imitation
(DASI) task.
 ARTICLE IN PRESS
6 Journal of Voice, Vol. ■■, No. ■■, 2017
FIGURE 3. Scatter plots of the relationships between the precision scores of pitch-matching from the pure tone imitation (PTI) task and the
self-advantage for accuracy (A) and precision (B). The types of mismapping were determined using the pitch-matching from the PTI task.
participants showed the best performance on the SASI task (ac-
curacy score = 26 cents; precision score = 38 cents), followed
by the DASI task (accuracy score = 43 cents; precision score = 60
cents), and finally the PTI task (accuracy score = 184 cents; pre-
cision score = 129 cents). The results suggested that the self-
advantage depends on the articulation consistency between the
target and the intended vocal action.
Criterion 2: dependence of phonation on the
articulation consistency between the target and the
intended vocal action
The dependence of phonation on the articulation consistency for
erroneous mapping, no mapping, and accurate mapping (the
control condition) was estimated using two measures: (1) the
size of the self-advantage (ie, the differences in accuracy scores
and precision scores between the DASI and the SASI) and (2)
the linear relationship between the self-advantage and the pre-
cision scores. This validity test was based on the pitch-matching
identified in the PTI task for the types of mismapping. The ac-
curacy and precision scores were analyzed using two mixed linear
models. Both of the models included the self-advantage as the
dependent variable, the scale degree (I–XII) and the type of
mismapping (accurate mapping, erroneous mapping, and no
mapping) as the fixed factors, the precision scores of pitch-
matching in the PTI task as the covariate, and the fixed and
random intercepts for the participants.
For the self-advantage for accuracy, there was no main effect
of type of mismapping (F[2,357] = 1.198, P = 0.303); however,
there was a significant interaction effect between precision scores
and the type of mismapping (F[3,362] = 4.323, P = 0.005). With
no mapping, the self-advantage was significantly correlated with
precision scores (b = 0.150, SE = 0.042, P < 0.001), but there was
no significant correlation between the self-advantage and the pre-
cision scores for pitch-matching with either erroneous mapping
or accurate mapping. Figure 3A shows the relationship between
the self-advantage for accuracy and precision of pitch-matching
with two types of mismapping.
For the self-advantage for precision, although there was a sig-
nificant main effect of the type of mismapping (F[2,358] = 4.363,
P = 0.013), multiple comparisons using the Bonferroni correc-
tion revealed no significant difference between each two pitch-
matching tasks. There was a significant interaction effect between
precision scores and the type of mismapping (F[3,363] = 11.305,
P < 0.001). With no mapping, the self-advantage was signifi-
cantly correlated with precision scores (b = 0.199, SE = 0.035,
P < 0.001). In contrast, there was no such correlation between
the self-advantage and the precision scores for pitch-matching
with either erroneous mapping or accurate mapping. Figure 3B
shows the relationship between the self-advantage for precision
and precision of pitch-matching with two types of mismapping.
Mismapping on a continuum
In the validity test presented above, we addressed mismapping
as a categorical variable. Since previous studies suggested that
categorization might not be the best way to define singing
deficiency,1,3,36 we then treated the mismapping as a continu-
ous variable to further test the proposed sensorimotor mismapping
hypothesis. In particular, we investigated the linear relation-
ship between precision scores and three criteria: (1) accuracy
scores; (2) the size of the self-advantage; and (3) the slope of
the regression line between precision scores and the magni-
tudes of self-advantage.
Criterion 1: accuracy
We performed a mixed linear model for each pitch-matching task.
In the model, we set the accuracy scores of pitch-matching as
the dependent variable, the scale degree as the fixed factor, and
the precision scores of pitch-matching as the covariate, and we
set the fixed and random intercepts for participants. All three of
the models showed significant main effects of precision (PTI:
F[1,371] = 62.467, P < 0.001; SASI: F[1,73] = 134.772, P < 0.001;
DASI: F[1,123] = 63.782, P < 0.001). There were significant pos-
itive correlations between precision scores and accuracy scores
on the PTI (b = 0.657, SE = 0.083, P < 0.001), the SASI
(b = 0.539, SE = 0.046, P < 0.001), and the DASI (b = 0.401,
SE = 0.050, P < 0.001).
Criterion 2: size of the self-imitation advantage
We created mixed linear models for the datasets for accuracy
scores and precision scores. Each model included the self-
advantage as the dependent variable, the scale degree (I–XII)
as the fixed factor, the precision scores on the PTI task as the
 ARTICLE IN PRESS
Hao He and Wei-dong Zhang Sensorimotor Mismapping in Pitch-matching
covariate, and the fixed and random intercepts for the partici-
pants. Both of the models revealed significant main effects of
precision on the datasets for accuracy scores (F[1,249] = 14.247,
P < 0.001) and precision scores (F[1,355] = 23.234, P < 0.001).
The precision of pitch-matching was significantly correlated with
the self-advantage for accuracy (b = 0.120, SE = 0.032, P < 0.001)
and precision (b = 0.136, SE = 0.028, P < 0.001).
Criterion 3: slope of the regression line between the
self-advantage and the precision of pitch-matching
For the 12 instances of pitch-matching for each target pitch height
of each participant, we computed the slope of the regression line
between the size of self-advantage and the precision scores. Con-
sequently, each participant had a datum of the slope for accuracy
and precision respectively. We then performed regression anal-
yses to investigate the linear relationship between the precision
scores of pitch-matching in the PTI task and the slopes. The results
showed that for accuracy, there was a significant correlation
between precision scores and the slopes (b = 0.002, SE = 0.001,
P = 0.042). For precision, the linear correlation between preci-
sion scores and the slopes was marginally significant (b = 0.001,
SE = 0.001, P = 0.055).
Testing collinearity
For the above validity test that examined the linear relation-
ship between precision scores and accuracy scores, a concern
arose regarding the possible collinearity between the two vari-
ables, as they were computed using the same data. We thus
performed Monte Carlo simulations to investigate whether the
correlation between precision and accuracy was a foregone con-
clusion. Given that the signed pitch deviations of pitch-matching
followed a Gaussian distribution,36 we used the Matlab func-
tion of “normrnd” to construct a Gaussian distribution with a
mean of zero and an SD of 100. We generated 100 random
samples. Each of them had 100 data points that simulated the
signed pitch deviations produced in the pitch-matching trials.
For each sample, we computed the mean of the signed pitch de-
viations and then determined the absolute value, and we computed
the SD of the signed pitch deviations. Afterward, we computed
the Pearson product-moment correlation coefficient of the two
variables. We ran the simulation 1000 times. The ratio of pos-
itive correlation coefficients that were significant was 0.017. This
result indicated that for a given dataset, the absolute value of
the mean and the SD had a low probability of being correlated.
This suggested that the observed correlation between the accu-
racy and the precision of pitch-matching could not be simply
attributed to the collinearity of the two measures.
DISCUSSION
The present study aimed to explore the construct of sensorimo-
tor mismapping in poor-pitch singing. In doing so, we attempted
to understand the properties, prevalence, and possible causes of
the two types of mismapping. Additionally, we attempted to in-
vestigate whether and to what extent the two types of mismapping
could be predicted by the MMIA model, which explained the
formation process of the sensorimotor mismapping. Finally, we
discussed the proposed and verified hypothesis with respect to
7
the mechanism of the self-imitation advantage in this study. This
discussion was organized around these three points.
Types of sensorimotor mismapping
The two types of sensorimotor mismapping, erroneous mapping
and no mapping, were operationally defined according to the pre-
cision and accuracy of pitch-matching. The present data showed
that the definitions are reliable and valid. As the validity tests
demonstrated, the two types of mismapping differed according
to two criteria: (1) the accuracy of pitch-matching and (2) the
dependence of phonation on the articulation consistency between
the target and the intended vocal action.
Specifically, for the first criterion (ie, accuracy), erroneous
mapping was associated with more accurate pitch-matching than
no mapping in the PTI task. Furthermore, there were signifi-
cant positive correlations between precision and accuracy in all
the three pitch-matching tasks. Moreover, the Monte Carlo sim-
ulation indicated that the observed correlation between the
accuracy and precision of pitch-matching could not simply be
attributed to the collinearity of the two measures. For the second
criterion, the dependence of phonation on the articulation con-
sistency between the target and the intended vocal action, the
reported results revealed a significant linear correlation between
the self-advantage and the precision scores in cases of no mapping
but not in cases of erroneous mapping. Moreover, there was no
such correlation in cases of accurate mapping, which served as
the control condition. The result was consistent with the pro-
posed hypothesis. Because phonation is guided by the
sensorimotor mapping formed in the erroneous mapping con-
dition and the accurate mapping condition, it is less dependent
on articulatory motor cues than phonation in the no mapping con-
dition. However, there was no significant difference in the
magnitude of the self-advantage between pitch-matching with
erroneous mapping and pitch-matching with no mapping. This
finding might arise from a floor effect caused by the similar dif-
ficulty levels of the two self-imitation tasks.
The validity testing in this study also considered current con-
cerns regarding categorizing pitch-singing proficiency. 1,36
Categorization has two limits: First, because categorization is
based on cutoffs, it is difficult to compare data from studies that
use different cutoffs; second, because there are usually remark-
able differences in performance between good and poor singers
or between easy and difficult tasks, the assumption of homo-
geneity of variance for ANOVA is likely to be violated, which
can produce biased results. To address these issues, research-
ers have suggested treating singing ability as a continuous variable
rather than a categorical variable.36 Technically, then, it would
be more appropriate to use regression-based analyses than mean
comparison methods, such as ANOVA. Our hypothesis sur-
vived the validity tests in which mismapping was investigated
on a continuum.
Regarding the prevalence of the two types of mismapping, the
focus was on the frequency of erroneous mapping because pitch-
matching with no mapping unquestionably accounts for the
majority of pitch-matching with mismapping because of the as-
sociation between accuracy and precision. In this study, the
percentage of pitch-matching with erroneous mapping among
 ARTICLE IN PRESS
8 Journal of Voice, Vol. ■■, No. ■■, 2017
all pitch-matching errors (including inaccurate pitch-matching
and imprecise pitch-matching) was 32% on the PTI, 70% on the
SASI, and 44% on the DASI. Approximately one-tenth of pitch-
matching was attributed to erroneous mapping when a more
stringent definition was used (ie, when both accuracy and pre-
cision cutoffs were 50 cents): 11% on the PTI, 7% on the SASI,
and 16% on the DASI. The data differed from that of a previ-
ous study in which inaccurate but precise pitch-matching was
rare (2%).2 The divergence between the two studies might be
explained by the following reasons. First, unlike the present study,
in which all of the participants were inaccurate singers, most
of the participants in that study were accurate singers, which led
to a low ratio of inaccurate pitch-matching and might have further
resulted in the rarity of inaccurate yet precise pitch-matching.
Second, the cutoff for accuracy was more liberal in the previ-
ous study (100 cents) than in the present study (50 cents). The
looser criterion decreased the number of instances of inaccu-
rate pitch-matching and contributed to the rarity of inaccurate
yet precise pitch-matching. Third, previous research showed that
pitch-matching was more precise in a single-pitch imitation task
than in a novel melody imitation task.2,3 In summary, the prev-
alence of erroneous mapping cannot be neglected, although it
accounts for a minor part of pitch-matching errors.
Regarding the causes of erroneous mapping, in addition to the
two reasons discussed in the “Hypothesis testing” section (ie,
the compromise strategy and the limited perceptual resolution
of human vocalization), a third cause may be the dependence
on comfort pitch. Individuals are likely to sing an easily pro-
duced tone when they have no idea how to imitate the target pitch.
The extreme case is “monotone” singing, in which individuals
produce almost the same pitch regardless of the target they are
attempting to match. More frequently, individuals have several
comfort pitches; therefore, the comfort pitch used may depend
on its closeness to the target. As previous research suggested,4,21
comfort pitch dependence reflects the contribution of vocal motor
control to pitch-matching. Because comfort pitches are easily
produced, the vocal motor actions used to produce them have
lower engagement thresholds. Thus, when there is no mapping
of a pitch percept, individuals are likely to connect the pitch
percept to a nearby phonatory representation with a higher pre-
activity level. Pitch-matching in such cases may be associated
with large pitch deviations.
Types of sensorimotor mismapping and the MMIA
model
The previous MMIA model, which provided a mathematical ex-
planation for sensorimotor translation in pitch singing, succeeded
in representing the properties of pitch-matching: the compres-
sion of pitch range, the attraction effect of comfort pitch, and
the presence of larger pitch deviations in single-pitch match-
ing than in interval or melody matching.19 In this study, we
performed simulations of the MMIA model to investigate whether
the observed pitch-matching that was attributed to types of
mismapping could be predicted by the model and to what extent
it could be predicted. In particular, we focused on how the pitch-
matching accuracy scores could vary with the variability of
FIGURE 4. Simulations of the multimodal imagery association
(MMIA) model representing the relationship between the accuracy scores
of pitch-matching and two sources of variability (SD = standard devi-
ation). See text for details.
mismapping (ie, the precision of pitch-matching) and the vari-
ability of bias that represented the attraction of comfort pitch.
There were four input parameters in the MMIA model: target
pitch, comfort pitch, SD of mapping (ie, SD of unbiased sung
pitch distribution), and SD of bias (ie, SD of comfort pitch
distribution).19 We used all the 12 degrees of a chromatic scale,
coded in cents as 100, 200, 300. . . 1200, as input parameters
for both target pitch and comfort pitch for the model because
all the degrees of a chromatic scale were applied in this study.
Twenty values, coded in cents as 10, 20, 30. . . 200, were used
as the input for SD of mapping. Eighteen values, coded in cents
as 30, 40, 50. . . 200, were used as the input for SD of bias. Two
hundred “trials” were run for each combination of the above four
input parameters.
Figure 4 illustrates the covarying relationship between accu-
racy scores and two sources of variability. The simulation showed
that the accuracy of pitch-matching decreased as the variabili-
ty of mismapping increased at each level of the variability of
bias. This result represented both pitch-matching with accurate
mapping and a large part of pitch-matching with no mapping.
Additionally, there were considerable instances of inaccurate
pitch-matching in which the SDs of mismapping were less than
100 cents, which represented erroneous mapping. Pitch-
matching with erroneous mapping was associated with a small
SD of bias, suggesting that erroneous mapping was more likely
to occur in individuals with a narrow comfort-pitch range than
those with wide range. When the comfort-pitch range is very
narrow, it leads to “monotone” singing, which generally has
large pitch deviations. However, as Figure 4 shows, the com-
bination of low variability of both mismapping and bias is
associated with accurate pitch-matching. Moreover, the simu-
lations of the MMIA model failed to represent imprecise yet
accurate pitch-matching.
In summary, in most cases, the MMIA model could predict
the types of mismapping proposed in the present study, but the
model was challenged at some points.
 ARTICLE IN PRESS
Hao He and Wei-dong Zhang Sensorimotor Mismapping in Pitch-matching
Self-imitation advantage
The self-imitation advantage played a crucial role in the valid-
ity test that used the criterion of the dependence of phonation
on the articulation consistency between the target and the in-
tended vocal action. Regarding the mechanism underlying the
self-imitation advantage, previous literature has attributed it to
timbre similarity.4,31 However, evidence shows that timbre sim-
ilarity cannot be the primary cause of the self-advantage.6 Other
researchers have explained the self-advantage in view of the mirror
neuron system, arguing that the auditory percept gives rise to
the activation of brain areas responsible for vocal motor planning.37
However, the mirror neuron system explanation of the self-
advantage is not sufficiently precise because it explains only the
initiation of the imitation response and does not assure accu-
rate imitation, especially in pitch-matching, in which phonation
cannot be directly observed. Moreover, the brain areas respon-
sible for vocal motor planning are activated even when individuals
listen to instrumental sounds without cues regarding human vocal
movements.38 Recent research has suggested that prior success-
ful auditory-motor experience plays a key role in the
self-advantage.6 Based on this finding, we offer a more de-
tailed explanation in view of the transfer-appropriate processing
model.39,40 In this regard, we view pitch-matching as a state-
dependent memory process and propose that the articulation
consistency between the target and the intended vocal action leads
to the self-advantage. The hypothesis was verified by the present
data. The self-advantage occurred between each two pitch-
matching tasks.
Although timbre is not a deeply rooted reason for the self-
advantage, it very likely plays a role in pitch-matching at the
operational level as some individuals may imitate pitch through
timbre matching. Indeed, subjective attributes such as pitch and
timbre are partly intertwined as they share physical attributes
of sound.41 Moreover, timbre largely depends on articulation,42–44
and articulation affects phonation.27–29 Therefore, pitch-matching
may occur through timbre matching. This conjecture may explain
why pitch-matching is better for synthesized tones with the ap-
proximate timbre of the human voice than for sine waves, is better
for the tones of human voices than for the tones of instrument
sounds,4,30–32 and is better for same-sex voices than for opposite-
sex voices.45 Even when repeating phonemes presented by
members of the opposite sex, participants unconsciously imitate
the voice spectra of the target.30,46 These results suggest that ar-
ticulation consistency and timbre similarity affect pitch-
matching and have implications for the treatment of poor-pitch
singing.
CONCLUSION
In the present study, we proposed a hypothesis that the senso-
rimotor mismapping in poor-pitch singing can be categorized
into two types: erroneous mapping and no mapping. Pitch-
matching with erroneous mapping is associated with good
precision, whereas pitch-matching with no mapping is associ-
ated with poor precision. We operationally defined the two types
of mismapping based on the precision of pitch-matching. Our
operational definitions were demonstrated to be reliable and valid.
As the criterion-related validity test showed, the two types of
9
mismapping differed in two aspects. First, pitch-matching with
erroneous mapping was more accurate than pitch-matching with
no mapping. Second, phonation was more dependent on the ar-
ticulation consistency between the target and the intended vocal
action in cases of no mapping than in cases of erroneous mapping.
These properties of mismapping were verified by further anal-
yses that addressed mismapping on a continuum. In addition,
the proposed types of mismapping could be predicted by the pre-
vious MMIA model. The contribution of the present study lies
in two aspects: First, we elaborated the construct of mismapping
of poor-pitch singing at a category level, and second, we pro-
posed and verified a further detailed explanation for the self-
imitation advantage in pitch-matching.
Acknowledgment
We thank Qinglin Meng and Zhenya Yang for their valuable com-
ments regarding the acoustic assessment.
REFERENCES
1. Dalla Bella S. Defining poor-pitch singing. Music Percept. 2015;32:272–282.
2. Pfordresher PQ, Brown S, Meier KM, et al. Imprecise singing is widespread.
J Acoust Soc Am. 2010;128:2182–2190.
3. Berkowska M, Dalla Bella S. Uncovering phenotypes of poor-pitch singing:
the Sung Performance Battery (SPB). Front Psychol. 2013;4:1–12.
4. Hutchins SM, Peretz I. A frog in your throat or in your ear? Searching for
the causes of poor singing. J Exp Psychol Gen. 2012;141:76–97.
5. Pfordresher PQ, Brown S. Poor-pitch singing in the absence of “tone
deafness.” Music Percept. 2007;25:95–115.
6. Pfordresher PQ, Mantell JT. Singing with yourself: evidence for an inverse
modeling account of poor-pitch singing. Cogn Psychol. 2014;70:31–57.
7. Zarate JM. The neural control of singing. Front Hum Neurosci. 2013;7:1–12.
8. Loui P. A dual-stream neuroanatomy of singing. Music Percept.
2015;32:232–241.
9. Berkowska M, Dalla Bella S. Acquired and congenital disorders of sung
performance: a review. Adv Cogn Psychol. 2009;5:69–83.
10. Dalla Bella S, Berkowska M, Sowinski J. Disorders of pitch production in
tone deafness. Front Psychol. 2011;2:1–11.
11. Pfordresher PQ, Mantell JT. Singing as a form of vocal imitation:
mechanisms and deficits. Proceedings of the 7th Triennial Conference of
European Society for the Cognitive Sciences of Music. Jyväskylä. 2009.
12. Amir O, Amir N, Kishon-Rabin L. The effect of superior auditory skills
on vocal accuracy. J Acoust Soc Am. 2003;113:1102.
13. Ayotte J, Peretz I, Hyde K. Congenital amusia: a group study of adults
afflicted with a music-specific disorder. Brain. 2002;125:238–251.
14. Dalla Bella S, Giguère JF, Peretz I. Singing proficiency in the general
population. J Acoust Soc Am. 2007;121:1182–1189.
15. Estis JM, Dean-Claytor A, Moore RE, et al. Pitch-matching accuracy in
trained singers and untrained individuals: the impact of musical interference
and noise. J Voice. 2011;25:173–180.
16. Moore RE, Keaton C, Watts C. The role of pitch memory in pitch
discrimination and pitch matching. J Voice. 2007;21:560–567.
17. Nikjeh DA, Lister JJ, Frisch SA. The relationship between pitch
discrimination and vocal production: comparison of vocal and instrumental
musicians. J Acoust Soc Am. 2009;125:328–338.
18. Watts C, Moore R, McCaghren K. The relationship between vocal pitch-
matching skills and pitch discrimination skills in untrained accurate and
inaccurate singers. J Voice. 2005;19:534–543.
19. Pfordresher PQ, Halpern AR, Greenspon EB. A mechanism for sensorimotor
translation in singing: The Multi-Modal Imagery Association (MMIA) model.
Music Percept. 2015;32:242–253.
20. Barlow CA, Howard DM. Voice source changes of child and adolescent
subjects undergoing singing training—a preliminary study. Logop Phoniatr
Vocology. 2009;27:66–73.
21. Joyner DR. The monotone problem. J Res Music Educ. 1969;17:115.
 ARTICLE IN PRESS
10
22. Estis JM, Coblentz JK, Moore RE. Effects of increasing time delays on
pitch-matching accuracy in trained singers and untrained individuals. J Voice.
2009;23:439–445.
23. Dalla Bella S, Giguère JF, Peretz I. Singing in congenital amusia. J Acoust
Soc Am. 2009;126:414–424.
24. Kawato M. Internal models for motor control and trajectory planning. Curr
Opin Neurobiol. 1999;9:718–727.
25. Cook DA, Beckman TJ. Current concepts in validity and reliability for
psychometric instruments: theory and application. Am J Med. 2006;119:166,
e7-16.
26. Kimberlin CL, Winterstein AG. Validity and reliability of measurement
instruments used in research. Am J Health Syst Pharm. 2008;65:2276–2284.
27. Erickson DM. The geniohyoid and the role of the strap muscles in pitch
control. J Acoust Soc Am. 1976;60:S63.
28. Honda K, Baer T, Hirose H, et al. Relationship between vowel articulation
and pitch control. J Acoust Soc Am. 1981;69:S67.
29. Sapir S, Campbell C, Larson C. Effect of geniohyoid, cricothyroid and
sternothyroid muscle stimulation on voice fundamental frequency of
electrically elicited phonation in rhesus macaque. Laryngoscope.
1981;91:457–468.
30. Granot RY, Israel-Kolatt R, Gilboa A, et al. Accuracy of pitch matching
significantly improved by live voice model. J Voice. 2013;27:390, e313-e320.
31. Moore RE, Estis J, Gordon-Hickey S, et al. Pitch discrimination and pitch
matching abilities with vocal and nonvocal stimuli. J Voice. 2008;22:399–407.
32. Price HE, Yarbrough C, Jones M, et al. Effects of male timbre, falsetto, and
sine-wave models on interval matching by inaccurate singers. J Res Music
Educ. 1994;42:269.
33. Grassi M, Soranzo A. MLP: a MATLAB toolbox for rapid and reliable
auditory threshold estimation. Behav Res Methods. 2009;41:20–28.
Journal of Voice, Vol. ■■, No. ■■, 2017
34. Lartillot O, Toiviainen P, Eerola T. A Matlab toolbox for music information
retrieval. In: Preisach C, Burkhardt H, Schmidt-Thieme L, et al., eds. Data
Analysis, Machine Learning and Applications. Berlin: Springer; 2008:261–
268.
35. Field A. Discovering Statistics Using SPSS. London: Sage Publications; 2009.
36. Pfordresher PQ, Larrouy-Maestri P. On drawing a line through the
spectrogram: how do we understand deficits of vocal pitch imitation? Front
Hum Neurosci. 2015;9:271.
37. Lévêque Y, Giovanni A, Schön D. Pitch-matching in poor singers: human
model advantage. J Voice. 2012;26:293–298.
38. Leaver AM, Van Lare J, Zielinski B, et al. Brain activation during anticipation
of sound sequences. J Neurosci. 2009;29:2477–2485.
39. Blaxton TA. Investigating dissociations among memory measures: support
for a transfer-appropriate processing framework. J Exp Psychol Learn
Memory Cogn. 1989;15:657–668.
40. Morris CD, Bransford JD, Franks JJ. Levels of processing versus transfer
appropriate processing. J Verbal Learn Verbal Behav. 1977;16:519–533.
41. Houtsma AJM. Pitch and timbre: definition, meaning and use. J New Music
Res. 1997;26:104–115.
42. Jürgens U. Neural pathways underlying vocal control. Neurosci Biobehav
Rev. 2002;26:235–258.
43. Sundberg J. The Science of the Singing Voice. DeKalb, IL: Northern Illinois
University Press; 1987.
44. Titze IR. Principles of Voice Production. Englewood Cliffs, NJ: Prentice
Hall; 1994.
45. Watts CR, Hall MD. Timbral influences on vocal pitch-matching accuracy.
Logopedics, phoniatrics, vocology. 2008;33:74–82.
46. Gentilucci M, Bernardis P. Imitation during phoneme production.
Neuropsychologia. 2007;45:608–615.