Content uploaded by Cara Stepp
Author content
All content in this area was uploaded by Cara Stepp
Content may be subject to copyright.
Content uploaded by Cara Stepp
Author content
All content in this area was uploaded by Cara Stepp
Content may be subject to copyright.
1
Surface electromyography for speech and swallowing systems:
Measurement, analysis, and interpretation
Cara E. Stepp
1§
1
Department of Speech, Language, & Hearing Sciences, Boston University, Boston, MA
§
Corresponding author, cstepp@bu.edu
Running Head: Surface Electromyography
Keywords: voice, speech, swallowing, respiration, electromyography
This is an author-produced manuscript that has been peer reviewed and accepted for publication in the Journal of Speech,
Language, and Hearing Research (JSLHR). As the “Papers in Press” version of the manuscript, it has not yet undergone
copyediting, proofreading, or other quality controls associated with final published articles. As the publisher and copyright
holder, the American Speech-Language-Hearing Association (ASHA) disclaims any liability resulting from use of inaccurate or
misleading data or information contained herein. Further, the authors have disclosed that permission has been obtained for use
of any copyrighted material and that, if applicable, conflicts of interest have been noted in the manuscript.
.http://jslhr.asha.orgThe final version is at
JSLHR Papers in Press. Published January 9, 2012, as doi: 10.1044/1092-4388(2011/11-0214)
Copyright 2012 by American Speech-Language-Hearing Association.
2
Abstract
Purpose: Applying surface electromyography (sEMG) to the study of voice, speech, and
swallowing is becoming increasingly popular. An improved understanding of sEMG and
building a consensus as to appropriate methodology will improve future research and clinical
applications.
Method: An updated review of the theory behind recording sEMG for the speech and swallowing
systems is provided. Several factors known to affect the content of the sEMG signal are
discussed, and practical guidelines for sEMG recording and analysis are presented, focusing on
special considerations within the context of the speech and swallowing anatomy.
Results: Unique challenges are seen in application of sEMG to the speech and swallowing
musculature owing to the small size of the muscles in relation to the sEMG detection volume and
the current lack of knowledge about innervation zone locations.
Conclusions: Despite the challenges discussed, application of sEMG to speech and swallowing
has potential as a clinical and research tool when used correctly, and is specifically suited to non-
invasive clinical studies utilizing between-condition or group comparisons for which detection of
specific isolated muscle activities is not necessary.
3
Introduction
There is an accumulating body of research utilizing surface electromyography (sEMG) for
assessment and rehabilitation of speech and swallowing disorders. sEMG has been used to study
and rehabilitate respiration and speech breathing (e.g., Maarsingh, Oud, van Eykern, Hoekstra, &
van Aalderen, 2006; Tamplin et al., 2011), voice (e.g., Allen, Bernstein, & Chait, 1991;
Andrews, Warner, & Stewart, 1986; Hocevar-Boltezar, Janko, & Zargi, 1998; Stemple, Weiler,
Whitehead, & Komray, 1980; Yiu, Verdolini, & Chow, 2005), swallowing (e.g., Crary &
Groher, 2000; Huckabee & Cannito, 1999), and speech articulation (e.g., Deng et al., 2009;
McClean & Tasko, 2003; Ruark & Moore, 1997). The attraction is clear – sEMG is non-
invasive, seemingly simple to apply, and can provide real-time information about muscle
activations. However, sEMG is a technique that can be easily abused due to a lack of knowledge
of the factors affecting the signal, inherent technical limitations (e.g., De Luca, 1997), and the
anatomy and physiology of the head and neck musculature. Lack of understanding of these
issues may explain the inconsistencies in clinical adoption of sEMG for assessment and
treatment of voice, speech, and swallowing.
For instance, sEMG signal differences could result from variations in recording methodologies.
Surface electrodes intended to detect muscle activation can be placed in nearly any configuration
on the body and still detect electrical activity of some kind, including cardiac activity and
electrical line noise. Several protocols have been developed for electrode placement to avoid the
potential signal misinterpretation and to increase detection reliability (e.g., De Luca, 1997;
Hermens, Freriks, Disselhorst-Klug, & Rau, 2000; Hermens et al., 1999). It is imperative that
both investigators and consumers of sEMG research understand the appropriate methodologies
4
and limitations: to avoid both wasting time with poor study design as well as misinterpreting
data, which could lead to reduced quality of patient care.
Electrophysiology and the technical aspects of sEMG recording methodology are not commonly
included in the standard educational preparation in the discipline of speech language pathology;
thus, expertise and experience in this area varies widely among researchers and clinicians in our
field. Improved understanding of sEMG and a consensus among our community as to
appropriate methodology will improve future research and clinical applications. Although
Cooper reviewed EMG for speech research in 1965 and Gay and Harris updated the review in
1971, both of these were primarily focused on needle and hooked wire EMG (Cooper, 1965; Gay
& Harris, 1971). Needle and hooked wire EMG are a viable source of information about muscle
activity during speech and swallowing, providing information that surface recordings cannot.
However, because of its noninvasive nature, sEMG is also now a methodology of choice for a
variety of clinical and research applications and provides complementary information to that
acquired through invasive techniques. Thus, this tutorial will provide an updated review of the
theory behind recording sEMG for the speech and swallowing systems. We first present details
about the generation of the underlying signal, we then review signal detection, the signal
composition, data analysis techniques, and possible forms of signal degradation. We close with a
focus on special considerations for speech and swallowing anatomy and future directions.
Motor Unit Physiology and EMG
During muscle contraction, nerve impulses from alpha motor neurons reach motor end plates at
the neuromuscular junction. These pulses cause all muscle fibers innervated by that motor
neuron’s axon to discharge nearly synchronously to create a motor unit action potential (MUAP).
MUAPs then propagate along all innervated muscle fibers, away from the motor end plate
5
longitudinally in both directions toward the end of the muscle fiber. Thus, the electric potential
field generated by the depolarization of the extrafusal fiber membranes is essentially an
amplified version of the alpha motor neuron activity. The electromyogram (EMG) is a
representation of this “myoelectricity” as detected at some distance (see Moritani, Stegeman, &
Merletti, 2004 for a more detailed review).
The tissues separating the EMG signal sources (depolarized zones of the muscle fibers) from the
EMG sensor are referred to as a volume conductor. The volume conductor consists of resting
muscles, subdermal fat, other soft tissues, and the skin. The volume conductor acts like a spatial
low-pass filter on the electrical potential distribution, smoothing each MUAP and decreasing the
amplitude of the signal. The distance between the EMG signal sources and the sensor or sensors
changes the qualities of the volume conductor and thus the effects of the spatial low-pass
filtering. Greater distances constitute lower signal amplitude and increased smoothing. The EMG
may be measured intramuscularly or at the surface of the skin (sEMG), yielding different
information based on the distance of the observation site from the active muscle fibers
immediately beneath the skin and whether or not other muscles are within the region either
immediately beneath the skin or beneath the target muscle. For surface detection particularly, the
effect of the separating tissues can become significant, with more than 1-2 cm of subdermal fat at
a site precluding the usefulness of sEMG (Merletti, Botter, Troiano, Merlo, & Minetto, 2009).
The sEMG signal is a collection of the multiple MUAPs within the range of the sensor,
providing a polyphasic signal of superimposed MUAPs from one or more muscles in the region.
The amplitude and frequency content of each of the constitutive MUAPs in the measured sEMG
signal is directly related to the distance of each motor unit from the electrode. A MUAP
measured from a more superficial muscle fiber will have a larger amplitude and higher frequency
6
content than one measured from a deeper muscle fiber (see Kamen & Caldwell, 1996 for a more
detailed discussion). As the central nervous system drives the muscle to generate increased force,
more motor units are recruited and the firing rates of all recruited motor units increase.
As indicated previously, the EMG signal can be detected from the surface of the skin (sEMG) or
through an inserted electrode (wire or needle). Intramuscular recordings achieved through the
use of needles or wires have the advantage of greater spatial and temporal specificity, and can
usually provide reliable information about the activation of a select number of motor units and
about the overall shape of individual MUAPs. However, intramuscular recordings are relatively
invasive, which could cause potential changes in behaviors of some disordered populations of
interest. In addition, because the information sampled comes from only a select few motor units,
it is not representative of the action of the entire muscle when using bipolar needle electrodes or
concentric needle electrodes. On the other hand, some intramuscular electrodes can record from
much larger regions within a muscle such as when there is a larger field between two hooked
wire electrodes or if a monopolar needle electrode and a distant ground is used. Also, depending
on the muscle of interest, it can be difficult to reliably place electrodes within the muscle body of
interest due to the lack of direct visualization. However, wire or needle electrodes are the only
way to measure EMG from deep muscles and can assure more selective recordings from single
muscles immediately beneath the skin surface.
For muscles that are relatively superficial, sEMG can be effective for detecting muscle
activation. Because sEMG detects signals from a larger area, the sEMG signal samples from
many motor units. This means that a sEMG signal may be representative of the overall activation
of the muscle of interest. However, because of the larger detection area, sEMG is also more
prone to detection of signals from nearby muscles (cross-talk) and conventional sEMG cannot
7
discriminate among regional differences in activation patterns within a muscle (e.g., Wentzel,
Konow, & German, 2010).
How to detect sEMG
Although sEMG can detect muscle activation, electrical potentials much larger than those
produced by muscles can contaminate the sEMG signal. Thus, sEMG signals should always be
collected relative to a common reference, referred to as the “ground.” Bioelectrical noise is
assumed to be common to both the ground and sensors allowing for the “common mode
voltages” to be rejected from the detected signal. The European Union sponsored a project
termed SENIAM (Surface Electromyography for the Noninvasive Assessment of Muscles) to
collect recommendations on sEMG methodology. In general, SENIAM suggests the wrist, the
spinous process of C7, or the ankle as appropriate locations for ground electrodes (Hermens et
al., 1999). For individuals recording sEMG from the muscles of speech and swallowing, we
recommend use of sites closer to sEMG sensors, such as the spinous process of C7, the acromion
process (bony prominence of the shoulder), forehead, nose, or earlobe. If monopolar electrodes
are used, care should be taken not to use a ground that is too low (i.e. with the heart placed
between the ground and the electrode) to minimize the effects of cardiac electrical activity in the
signal.
The ability to measure high-fidelity sEMG is dependent upon the impedance of the skin-
electrode interface (the opposition of current flow from the skin to the sensor). Lower impedance
of the interface corresponds to improved signal propagation to the electrode and better signal
detection. In its natural state, the top layer of the epidermis is electrically insulated, resulting in
high skin-electrode impedance. Depending on electrode characteristics and skin state, contact
impedance ranges from a few kΩ to a few MΩ, with larger electrodes generally having lower
8
impedance and noise (Merletti et al., 2009). This impedance can be reduced using a variety of
methods. The first choice of the researcher is whether to use passive or active electrodes.
Passive electrodes are made of conductive materials that sense electrical current on the skin
through the skin-electrode interface, the most simple of which are made of silver. Silver-silver
chloride electrodes are also used. These electrodes allow a reversible chloride exchange interface
between the electrode and skin and help to minimize motion artifact produced by skin potentials
(Webster, 1984). Passive electrodes are often referred to as “wet electrodes” as they require
conductive gel or paste between the electrode and skin to improve the quality of the detected
signal. However, Roy and colleagues have shown that the use of conductive gel in the face of
perspiration and mechanical perturbations can lead to an increase in the artifacts measured (Roy
et al., 2007).
Active electrodes are also referred to as “dry electrodes” or “pre-amplified electrodes.” These
electrodes have signal amplification circuitry embedded near the skin-electrode interface. Active
electrodes can increase the signal-to-noise ratio by minimizing source and contact noise, can be
used in situations with otherwise unacceptably high skin-electrode impedances, and do not
require the use of a conductive gel or paste. These electrodes are preferred in terms of signal
quality, but are often more expensive and more bulky than passive electrodes.
Regardless of the choice of electrode type, the signal detected can be improved by further
reducing the electrode-skin impedance through treatment of the skin. Techniques to remove the
dead (top) surface layer of skin and its protective oils can enhance electrode-skin contact,
resulting in a reduction of artifacts and noise. SENIAM recommends shaving the skin surface if
it is covered with hair, and cleaning the skin in question with alcohol (Hermens et al., 1999).
Although alcohol treatment has been recommended and has been widely adopted by clinicians
9
and researchers, this practice has been shown to reduce electrode-skin impedance by only
roughly 40% (Merletti & Hermens, 2004). Rubbing the skin with medical abrasive paste causes
the greatest reduction (roughly 90%) in electrode-skin impedance (Merletti & Hermens, 2004).
Because abrasive paste is often unpleasant due to messiness and discomfort, combining the use
of alcohol with skin “peeling” is suggested as a less invasive compromise. The practice of light
skin abrasion or “peeling” with adhesive tape is known to reduce electrode-skin impedance by
roughly 70% (Merletti & Hermens, 2004), and we have found it to be well-tolerated by
participants, even on delicate skin of the neck and face. It involves repetitive placement and
removal of adhesive tape on the skin surface.
As previously discussed, all signals should be recorded relative to the ground. However, in
addition, various electrode configurations can be used to further isolate electrical activity of
interest. A single electrode placed over a muscle and recorded relative to ground is referred to as
a monopolar configuration. Monopolar configurations are associated with the largest detection
volume and are most susceptible to crosstalk from adjacent muscles.
In order to remove interference sources and to compensate for the low-pass filtering effect of the
tissue, surface signals are typically detected using a linear combination of different electrodes,
the simplest of which is a differential electrode (Farina, Merletti, & Stegeman, 2004).
Differential recording configurations amplify the difference between multiple electrodes placed
over the muscle of interest. The most common of these is the bipolar configuration (single
differential). Differential recordings take advantage of “common-mode” rejection, such that
potential noise (biological or otherwise) that is sensed at both electrodes is rejected from the
amplified signal. In addition, differential recording configurations are more spatially sensitive
(smaller detection volume) than monopolar schemes. Double differential recording strategies
10
refer to three electrodes linearly configured over the muscle of interest, with three differences
among the electrodes used for the resultant signal. This configuration results in a further increase
in spatial selectivity relative to bipolar configurations (Merletti & Hermens, 2004). Regardless of
wehther the differential configuration is single or double, differential electrodes should be
aligned so that the electrode axis is parallel to underlying muscular fibers, detecting MUAPs as
they travel down the muscle fibers. When electrodes are not aligned parallel to muscle fibers, the
amplitude of the detected signal can be reduced by as much as 50% (Vigreux, Cnockaert, &
Pertuzon, 1979).
Features of differential sEMG depend upon on the size of and space between the electrodes,
referred to as the inter-electrode distance (Roeleveld, Stegeman, Vingerhoets, & Van Oosterom,
1997). SENIAM recommends a maximum electrode size of 10 mm in the muscle fiber direction,
with an interelectrode distance of approximately ¼ the length of the muscle fiber or 20 mm,
whichever is smaller (Hermens et al., 1999). For speech musculature, ¼ the length of the muscle
fiber is often smaller than 20 mm. For instance, muscles such as the sternocleidomastoid can be
roughly 20 cm in length, so 20 mm is easily a smaller distance than 5 cm. However many
muscles such as the mentalis and depressor labii inferior could be 2-4 cm in length, which would
require maximum electrode distances of 0.5 – 1 cm.
In general, the larger the inter-electrode distance, the wider the area sampled and the higher the
amplitude of the resultant signal, but the less spatially specific (Fuglevand, Winter, Patla, &
Stashuk, 1992; Roeleveld et al., 1997). Simulation has shown that larger inter-electrode distances
can moderately increase detection depth, but the detected sEMG signal is dominated by MUAPs
from muscle fibers located within 10-12 mm of the recording electrode (Fuglevand et al., 1992).
Further simulation has shown differences in the relationship between inter-electrode distances
11
and detection amplitude for superficial and deep fibers (Farina, Cescon, & Merletti, 2002). For
superficial fibers, the amplitude detected at the surface is reduced when the inter-electrode
distance is less than 15mm, whereas this cut-off is at 25mm for deeper fibers (Farina et al.,
2002). Although increases in the inter-electrode distance may increase the activity detected from
deeper fibers, activity from superficial fibers will still dominate the signal. Although simulation
has indicated that the size of electrodes used does not cause substantial effects on the detection
volume (Fuglevand et al., 1992), it has been asserted that smaller electrodes (diameter less than
5mm) are preferred for sEMG, as the larger electrodes introduce temporal low-pass filtering
(Merletti & Hermens, 2004).
sEMG signal composition and recording recommendations
Recording EMG requires preamplification of the signal, hardware or software filters (hardware
recommended), and an analog to digital (A/D) converter. Preamplification amplifies the original
detected signal prior to A/D conversion. This is done to maximize the fidelity of the recorded
signal. In active electrodes, preamplification takes place directly at the electrode head, allowing
amplification to take place before the possible introduction of noise through electrical cables.
The majority of the power of a typical sEMG signal is in the frequency range 0 - 450 Hz
(Merletti & Hermens, 2004). The remaining power at higher frequencies is mostly electrode and
equipment noise, the sources of which will be described in more detail in later sections (e.g., see
Sources of Noise). For this reason, it is common to filter the sEMG signal prior to A/D
conversion. Movement artifacts create signals in the 0 – 20 Hz range, and can be attenuated by a
high pass filter with a cut-off around 10 – 20 Hz (De Luca, Gilmore, Kuznetsov, & Roy, 2010).
Additionally, the sEMG signal should be low-pass filtered with a cutoff point around 500 Hz, to
remove high frequency noise. Investigators differ in their opinions as to whether this filtering
12
should be accomplished prior to digitization using hardware or post-digitization in software.
Hardware filtering is preferred by some as it allows the signal to be digitally acquired without the
possibility of aliasing. Software filtering is preferred by others because it allows investigators to
see the raw data, which may allow them to identify possible signal contamination that filtering
may hide.
Given analog low-pass filtering with a cut-off of roughly 500 Hz (implemented prior to
digitization), EMG data should theoretically be acquired with a minimum sampling rate of 1000
Hz to prevent aliasing. Aliasing occurs when an analog signal is under-sampled, such that the
reconstructed digital signal is distorted by high-frequency information from the original signal
occurring in the low frequencies of the digitally acquired signal. Aliasing can be prevented by
using a sampling frequency at least 2 times greater than the highest frequency present in the
original signal. Analog filtering prior to digitization allows attenuation of unwanted frequencies
(>500Hz), effectively preventing aliasing for appropriate sampling frequencies. Analog filter cut-
offs indicate a transition between the passband and stopband, but they are not infinitely steep:
even after filtering, some energy may remain in the signal at higher frequencies. Thus, to ensure
signal integrity of sEMG low-pass filtered at 500 Hz, data acquisition of at least 2000 Hz is
recommended. If low-pass filtering is to be accomplished after digitization, the risk of aliasing
high frequency noise to signal-bearing lower frequencies is vastly increased. Oversampling can
be used to avoid aliasing; this refers to the technique of using sampling frequencies many times
higher than the hypothetical Nyquist frequency. For instance, given that the majority of energy in
sEMG is less than 500 Hz, investigators may choose to oversample the signal by sampling at
10,000 Hz. However, other electrical activity other than sEMG may be present in the signal, such
that the total frequency profile in the measured signal is usually not under experimental control.
13
For this reason, digitization prior to anti-aliasing filtering is always risky, even with
oversampling, and therefore not recommended.
sEMG Signal Parameters
The raw acquired sEMG signal is often referred to as the interference sEMG or the interference
pattern. The interference sEMG can be used itself to provide gross information about muscle
activity; however, a variety of parameters can be estimated from the raw signal that are
commonly used to gain more reliable and specific information.
Amplitude Estimation
The overall amplitude of sEMG is of interest to researchers as the sEMG amplitude generally
increases with increases in muscle activation and force. However, this relationship is typically
not linear and is affected by a number of factors such as muscle length and fatigue (Disselhorst-
Klug, Schmitz-Rode, & Rau, 2009).
Although observation of interference sEMG can provide information about the amplitude of
sEMG, the peak amplitudes seen in the raw signal should not be used to estimate amplitude
because they can be due to a single motor unit and may not be representative of the overall
activity in the detection area. Commonly used amplitude estimators are the Average Rectified
Value (ARV) and Root Mean Square (RMS; see Equation 1). The sEMG signal is generally
well-described by a Gaussian distribution, leading to the assumption that the preferred estimator
of sEMG amplitude may be RMS, which has a smaller variance when predicting amplitude for
Gaussian distributions (Clancy & Hogan, 1999). However, results using experimental data
indicate that the ARV may provide smaller variance for amplitude estimation (Clancy & Hogan,
1999). Both techniques require the use of time windowing. When choosing a time window, there
14
is a trade-off between time-sensitivity and reliability of the estimate; smaller windows are more
sensitive to rapid changes in the signal, but inherently result in less reliable amplitude estimates.
SENIAM recommends windows of 250 – 500 ms for contraction levels above 50% maximal
voluntary contraction (MVC), or 1000 – 2000 ms for contraction levels below 50% MVC for
amplitude estimation. These window lengths are far too great to provide information about
dynamic movement. However, previous work by Norman and colleagues examined the
coefficient of variation of sEMG of the biceps integrated over varying window lengths, and
found that integration times greater than 75 ms resulted in greater reliability (Norman, Nelson, &
Cavanagh, 1978). Thus, optimal window lengths will vary based on task, and investigators must
choose a compromise between time-sensitivity and the quality of the amplitude estimate. Speech
articulatory movements tend to be fast with brief bursts and interburst intervals; thus, depending
on the specific task and the goal of the amplitude estimation, window lengths of 75 – 1000 ms
are generally recommended. Before any amplitude estimation technique is applied, investigators
should remove any DC offset from the signal such that the mean of the raw signal is zero.
N
i
i
x
N
RMS
1
2
1
Equation 1
Frequency Content
The frequency content of the sEMG signal is dependent upon the individual frequency
characteristics of the constitutive MUAPs as well as their relative distance from the recording
electrodes. Individual MUAPs have a specific size and shape, which determine their frequency
characteristics. Also, as the recruitment of the motor units changes, rate coding determines the
number of MUAPs per second. As muscle excitation increases, more individual MUAPs will
15
sum due to a larger number of recruited motor units and an increase in the rate of MUAPs from
each individual motor unit. The frequency content of the EMG is affected by both factors.
If individual MUAPs can be detected from the sEMG signal, one parameter of interest is the
firing rate of individual MUAPs, and the mean value of the firing rates of the MUAPs detected
(Basmajian, 1978). The frequency spectrum of sEMG may also provide information about
muscle activation. One of the most commonly used frequency-based measures is the median
frequency. The median frequency of the sEMG signal has been shown to strongly correlate with
localized muscle fatigue in a variety of physiological systems (e.g., Lindstrom, Kadefors, &
Petersen, 1977), and has been assessed in the speech/voice system using intramuscular EMG
(Boucher, Ahmarani, & Ayad, 2006) and sEMG of facial musculature (van Boxtel, Goudswaard,
van der Molen, & van den Bosch, 1983). The median frequency is the point at which the spectral
power of the signal is equally divided into low and high frequency halves. As muscles fatigue,
the median frequency of the power spectrum shifts to lower frequencies.
Although it is affected by the degree of volume conduction and size of muscle fibers, frequency
content is related to firing rate, and can be used to assess information about the co-activation of
multiple muscles. Muscle is thought to be driven by a number of different physiological
oscillations at varying frequencies (see Grosse, Cassidy, & Brown, 2002 for review), and these
oscillations may be characteristic of the function of distinct neural circuits. Frequency bands
such as alpha (8 – 13 Hz), beta (15 – 35 Hz), and gamma (30 – 70 Hz) have hypothesized
sources within the central nervous system. Coherence is a frequency domain measure of the
linear dependency or strength of coupling between two processes (e.g., Halliday et al., 1995),
and can be used to capture these physiological oscillations. The coherence function,
2
xy
R
,
16
can be defined as in equation 2 below, where f
xx
represents the auto-spectra of a time series x(t),
f
yy
the auto-spectra of y(t), f
xy
the cross-spectra x(t) and y(t), and λ the frequency of interest.
yyxx
xy
xy
ff
f
R
2
2
equation 2
Coherence between multiple EMG signals (intermuscular coherence) can be used to measure the
common presynaptic drive to motor neurons (Brown, Farmer, Halliday, Marsden, & Rosenberg,
1999), but has not yet been widely adopted in speech research. Smith and Denny have provided
the most information about intermusuclar coherence during speech production (Denny & Smith,
1992, 2000; Smith & Denny, 1990), characterizing the activation of the chest wall and masseter
muscles during a variety of speech, chewing, and breathing tasks (Smith & Denny, 1990). They
further examined the intermuscular coherence of lip, jaw, chest wall, and anterior neck
musculature of persons who stutter during speech and speech breathing (Denny & Smith, 1992,
2000). Goffman and Smith further investigated coherence between different quadrants of the
perioral region, finding a lack of functional coupling during speech and chewing tasks (Goffman
& Smith, 1994). More recently, anterior neck intermuscular coherence has been used in the study
of hyperfunctional voice production (Stepp, Hillman, & Heaton, 2010, 2011b).
Raw versus normalized amplitude
As mentioned, the tissues separating signal generation and signal detection have the effect of
low-pass filtering the sEMG signal, such that increases in the separation result in detection of a
smoothed signal with lower amplitude (Farina & Rainoldi, 1999). Increases in so-called skinfold
thickness can decrease the selectivity of the sEMG signal and results in more rapidly attenuated
signals (De la Barrera & Milner, 1994). In fact, models of varying levels of complexity have
17
shown that a majority of the amplitude of the sEMG signal can be lost with skinfold thicknesses
from 0.1 – 10 mm (e.g., Andreassen & Rosenfalck, 1978; Kuiken, Lowery, & Stoykov, 2003).
Thus, for some body areas of some participants, signal amplitudes are effectively too small to
measure with reasonable signal-to-noise ratios. Calipers should be used to determine fat layer
thickness of participants to ensure that appropriately strong signals can be measured.
Because small differences in submental fat can so greatly change the amplitude of the sEMG
signal measured, the raw amplitude of the signal is not a reliable measure among multiple
participants or even as a function of time in a single participants. Thus, sEMG signals should be
normalized to some reference contraction before they are compared between conditions and/or
participants to reduce the variability caused by differences in surface electrode contact and
submental fat levels (Netto & Burnett, 2006). Common references include MVC or some
percentage of the MVC (usually 50% or 60%). Studies have shown that sub-maximal
contractions are more reliable for simple, one-joint systems (Allison, Marshall, & Singer, 1993;
Yang & Winter, 1983). However, Netto and Burnett (2006) found that for anterior neck
musculature, the MVC reference was more reliable both within-day and between-days, and
speculated that this is likely due to the complex structure and synergistic action of neck
musculature. Reference contraction recommendations for muscles of the face and thorax are not
currently known.
Unfortunately, identifying tasks to induce MVC for speech musculature is not straightforward. In
the anterior neck, maximal neck contraction against manual resistance measured by a
dynamometer fit with a chin guard has been used with success (Stepp et al., 2011a). Manual
resistance of this type is not as well-suited for some other speech musculature such as the
18
muscles of respiration, for which more physiological tasks should be devised to prompt MVC
behaviors.
MVC normalization at best reduces the effects of differences in skin fold thickness. Great care
should be taken when comparing measures of sEMG amplitude across populations or even
within single subjects as a function of time, because normalization itself may introduce
uncertainty due to a lack of reliability in the reference contraction itself: participant perception
and production of maximal effort is easily affected by their environment and other motivating
factors. However, it is imperative to attempt proper normalization prior to making comparisons
of sEMG amplitudes in order for amplitude estimates to have physiological meaning.
Potential sources of sEMG signal degradation
Sources of Noise
sEMG signals can be degraded by several types of noise. The following section will review the
most common sources of noise, how to recognize the noise sources, and how to minimize their
effects. When recording sEMG, signals should always be monitored in real-time to ensure signal
integrity.
Perhaps the most common source of noise in the sEMG signal is power line interference. Power
line interference is noise resulting from the alternating current used to power electrical devices,
and is primarily at the line frequency of either 50 or 60 Hz, plus associated harmonics (at integer
multiples of the line frequency). This noise is typically larger than the sEMG signal in
magnitude. Power line interference in the sEMG signal can be vastly reduced by using active
electrodes, proper grounding, differential recording configurations, and properly shielded cables.
19
In addition, performing experiments in environments with limited electrical noise can also
reduce the effects of electrical line noise.
Poor signal integrity (low signal-to-noise ratios) due to high levels of subdermal fat can
exacerbate contamination of the signal with electrical line noise. As an example, Figure 2 shows
sEMG collected from the anterior neck during speech production from a young individual with
minimal subdermal neck fat and from a middle-aged participant with significant levels of
subdermal fat. Although troublesome, power line interference is typically easy to identify, both
in the time and frequency domain (see Figure 2). Post-processing with notch filters to remove 50
or 60 Hz interference is not advisable, as the power density of the sEMG signal is high in this
range, and the associated phase rotation can be introduced to the time waveform (Hermens et al.,
1999).
Low frequency artifact can contaminate the sEMG signal as a result of movement between the
electrode and skin or as a result of cable sway. Cable sway is less of an issue with active
electrodes because the signal is amplified prior to entering the cable. Movement artifacts can be
reduced by maintaining good adhesion of electrodes, using appropriate skin preparation
techniques, and high-pass filtering. Comprehensive testing of the relationship between high-pass
filter corner frequency and movement artifact has resulted in the general recommendation of a
Butterworth filter with a corner frequency of 20 Hz with a 12dB/oct slope as for general use (De
Luca et al., 2010). This result is consistent with the only work in this area specifically utilizing
speech musculature, which suggested optimal high pass cut-off frequencies between 15 – 25 Hz
(van Boxtel, 2001).
In brief, the power spectrum of the sEMG signal has a specific and fairly consistent profile,
whereas sEMG that has been contaminated with noise from motion artifacts and power line
20
interference shows distinct changes to that power spectrum. Even if all precautions for proper
recording are attempted and online monitoring of the time waveform does not reveal
contamination, before interpreting a signal, one should check for data quality by computing the
spectrum.
Effects of Innervation Zones
Neuromuscular junctions are often concentrated in a strip referred to as the innervation zone (or
active zone). Although the innervation zone is often idealized as a single strip, many muscles
have decentralized (diffusely localized) neuromuscular junctions. It is generally recommended
that differential electrodes be applied between the innervation zone and a tendon. In the past,
sensors have been instead placed over the center of the muscle (the belly) or over the innervation
zone (motor end plate zone), as this was the best location to record "large" monopolar sEMG
signals. It is now known that this location is not suitable for differential recordings; it is not
stable or reproducible because relatively small displacements of the sensors with respect to the
innervation zone cause large effects on the amplitude of the sEMG signal (Merletti & Hermens,
2004), particularly on the low-frequency (<110 Hz) components (Beck et al., 2009). This is
because the MUAPs propagate in both directions away from the innervation zone toward the
tendons. If bipolar electrodes are arranged on each side of the innervation zone, there is likely to
be a substantial level of cancellation of common signals, reducing the amplitude of the recorded
signal. In fact, differences in placement along the body of a muscle have been known to mask
differences in task conditions (Mercer, Bezodis, DeLion, Zachry, & Rubley, 2006). Thus, for
sEMG signals to be accurate and repeatable as possible, there must be a clear definition of
electrode position relative to the innervation zones (Hermens et al., 1999). When the locations of
21
innervation zones are unknown, use of double differential electrodes can reduce the effects of an
ill-placed sensor (Farina, Merletti, & Disselhorst-Klug, 2004).
Ideal sEMG recording procedures would first identify the innervation zones and find the optimal
electrode position on a subject by subject basis, using multi-channel electrode arrays. Falla,
Dall'Alba, Rainoldi, Merletti, and Jull (2002), for example, examined the sternocleidomastoid
(SCM) muscles in this way in 11 healthy normal individuals in order to determine specific
recommendations for electrode placement to optimize sEMG recordings (Falla et al., 2002).
Recommendations of this type are not currently available for most speech musculature, but some
muscles have been studied. One group has examined innervation zone locations for the muscles
of jaw elevation (Castroflorio et al., 2005); they found a large variability within and between
participants in the location of major innervation zones of the masseter, and could not recommend
an optimal position for electrode placement. Conversely, Lapatki and colleagues found distinct
clusters of motor endplates in the depressor anguli oris, depressor labii inferioris, mentalis, and
orbicularis oris inferior muscles across participants when topographical locations were spatially
warped to correct for anatomical differences between participants (Lapatki et al., 2006).
Similarly, the geniohyoid,has also been found to have clustered motor endplates, but with
distinct clusters located in separate compartments (Mu & Sanders, 1998).
When examining speech musculature, it may not be possible to completely avoid the effects of
innervation zones. However, researchers and clinicians should be aware of the possible effects of
innervation zones on the resulting signal, and should take steps to avoid them when possible,
such as using double differential electrodes.
Crosstalk
22
Although researchers and clinicians are often interested in detecting activation from isolated
muscles, muscles typically do not act in isolation. In the case of speech musculature, the
detection volume of sEMG is often large enough to detect activity from more than one muscle at
once, which is referred to as crosstalk. The sensor will detect from the nearest muscle as well as
those adjacent to it. When multiple signal sources are available, the sEMG interference pattern
will contain elements from all sources with the largest amplitudes coming from the sources
closest to the sensor.
The effects of crosstalk are minimized by decreasing the detection volume. This can be
accomplished by using smaller electrodes with smaller inter-electrode distances and double-
differential configurations (Koh & Grabiner, 1992, 1993). In the case of recording speech
musculature with sEMG, cross-talk may be unavoidable. If isolated muscle activations are of
interest, sEMG may not be appropriate. Unfortunately, due to the small size of some muscles and
regional differences in activation, even signals recorded using intramuscular electrodes can
contain crosstalk (e.g., Blair & Smith, 1986).
Special Considerations for Speech and Swallowing Anatomy
Scientists and clinicians wishing to measure sEMG from speech and swallowing anatomy have
special obstacles to overcome in data collection. These muscles are often small and have
overlapping fibers. Thus, it is often not possible to isolate the activity of single muscles of speech
and swallowing. However, sEMG can still provide a valuable tool to better understand and assess
speech and swallowing physiology as long as the limitations in recording are well understood
prior to interpretation of data. The following sections review work that has been accomplished in
the speech and swallowing anatomy that may guide investigators with specific recommendations
for electrode placement, confirmatory tasks, and signal interpretation.
23
Orofacial Musculature (muscles of articulation and mastication)
Despite their importance in understanding typical and disordered speech motor control, isolated
activations of orofacial musculature are difficult to acquire. A notable exception to this issue
would be the masseter muscle, which is large, simple to palpate, and located superficially. The
masseter can be easily recorded using sEMG and shows comparable results when studied
simultaneously by sEMG and intramuscular EMG (Koole, de Jongh, & Boering, 1991). In fact,
both amplitude and frequency parameters of sEMG of the masseter have been shown to be
reliable over multiple days (Suvinen, Malmberg, Forster, & Kemppainen, 2009). Similarly,
bipolar sEMG over the zygomaticus major region found significant correlations in sEMG
amplitude during specific facial poses recorded on different days (Tassinary, Cacioppo, & Geen,
1989).
The muscles of the lower face and submental area represent a greater challenge due to their small
size and overlapping fibers. For instance, surface signals detected over the anterior digastric
show differences in activation during isometric contractions from intramuscular recordings
(Koole et al., 1991). However, work from Lapatki and colleagues has provided guidance for the
muscles of the lower face: distinct clusters of motor endplates and primary muscle fiber
orientations have been shown in the depressor anguli oris, depressor labii inferioris, mentalis,
and orbicularis oris inferior muscles, which can lead to more informed placement of electrodes
(Lapatki et al., 2006). In their subsequent work they have used their methods to suggest optimal
placements of bipolar electrodes for muscles of the lower face: depressor anguli oris, depressor
labii inferioris, mentalis, and orbicularis oris inferior (Lapatki et al., 2010).
The work of Blair and Smith has shown that single muscles of human lips cannot be measured
with current technologies because of interdigitation of muscle fibers (Blair & Smith, 1986). They
24
concluded that this limitation of recording quality should be acknowledged during interpretation
of recordings. Despite this limitation, application of sEMG to the perioral region has resulted in a
substantial body of knowledge about the motor control of the muscles of the lips (e.g., Goffman
& Smith, 1994; Wohlert & Goffman, 1994; Wohlert & Hammen, 2000). In fact, in this region,
sEMG may offer as much discrimination as intramuscular electrodes. For instance, Lapatki and
colleagues have developed a small sEMG electrode for use with facial musculature, which has
been shown to have similar muscle selectivity as intramuscular recording techniques of the
muscles of the lower face (Lapatki, Stegeman, & Jonas, 2003).
Understanding what tasks can activate specific musculature can aid in confirmation of an
appropriate electrode position. Placement of electrodes over muscles of mastication such as the
buccinators and masseter can be aided by asking participants to clench their teeth and then
palpating to find the muscle body. Other smaller muscles can be more difficult. O’Dwyer and
colleagues have suggested procedures for the verification of hooked wire electrode placement for
a variety of orofacial and mandibular muscles, including gestures used as stimuli for
confirmation of activation (O'Dwyer, Quinn, Guitar, Andrews, & Neilson, 1981). The speech-
relevant muscles that could be detected with sEMG with satisfactory gestures for confirmation
based on O'Dwyer et al. (1981) and Burnett, Mann, Cornell, and Ludlow (2003) are shown in
Table 1. The gestures are known to elicit strong activation from each muscle, although not
isolated activation.
Submental and Anterior Neck Musculature
It is relatively easy to record from infrahyoid musculature using sEMG due to the prominent size
and superficial location of the sternohyoid and omohyoid muscles. High quality recordings can
be obtained from the neck surface by placing single or double differential sEMG electrodes 1 cm
25
lateral to the neck midline, and located from the gap between the cricoid and thyroid cartilages of
the larynx and as far superior as the border of the submental surface. Varying the superior-
posterior positions of electrodes can lead to some variation in the activity recorded. With a
surface electrode placed directly over the gap between the cricoid and thyroid cartilages,
previous authors have hypothesized that is possible to record from the cricothyroid, an intrinsic
laryngeal muscle; however, based on its relatively deep location it is unlikely to contribute to
surface recordings and past work using simultaneous intramuscular EMG and neck sEMG has
shown that surface recordings do not show evidence of CT activation (Loucks, Poletto, Saxon, &
Ludlow, 2005). Signals detected at this location are likely largely comprised of activations of the
sternohyoid and researchers interested in cricothyroid activation should use intermuscular EMG.
More superior placements are more likely to also include activations of the omohyoid.
There has been considerable interest in recording from the sternohyoid muscle for voice and
swallowing applications; however, the electrode locations and configurations employed have not
always been optimized. One suboptimal configuration that has been employed is to use a bipolar
recording configuration with each electrode located on opposite sides of the neck. This
configuration results in a recording of the difference in activation between the two sides, which
is essentially noise in the bilateral activation pattern. If a difference between the two sides is of
interest, this should be examined after differential signals from the two sides have been recorded.
The bipolar configuration has been developed to record reliably when electrodes are placed
longitudinally to the body of the muscle (see example configurations in Figure 3). If the
sternohyoid is of particular interest, spatial selectivity can be improved by using a double
differential electrode, which limits the contribution of deeper muscles to the detected signal.
26
Suprahyoid and submental musculature represent more of a recording challenge for using sEMG.
When measuring from the anterior neck, the thyrohyoid is deep to the sternohyoid and thus likely
does not contribute substantially to surface recordings. Recordings of the submental surface have
the potential to detect activation of the mylohyoid and the anterior belly of the digastric, but are
more difficult in many individuals due to increased subdermal fat in this area. Another issue is
the much smaller size of muscle bellies in this area and the overlapping fibers. Table 1 indicates
gestures known to elicit strong activation of many of the submental muscles.
Although it is arguably of limited importance during speech, the sternocleidomastoid is an
accessory respiratory muscle, and is known to activate during speech and singing (Pettersen,
Bjorkoy, Torp, & Westgaard, 2005). Due to its large size and superficial location, there is a large
body of previous research on recording sEMG from the sternocleidomastoid. In particular,
previous research has identified the common location of innervation zones in the SCM, and
recommended that electrodes should be placed 1/3 of the distance from the sternal notch to the
mastoid process, in the direction of the line from the sternal notch to the mastoid process in order
to avoid placement near innervation zones (Falla et al., 2002).
A particular issue for individuals interested in measuring from the submental surface and anterior
neck is contamination from the platysma. The platysma is a superficially located thin sheet of
muscle in the subcutaneous tissue of the neck. It extends over the anterolateral aspect of the neck
from the inferior border of the mandible to the superior aspect of the pectoralis major. Although
the activation of the platysma during speech has been studied somewhat less than other laryngeal
and orofacial musculature, it is thought to be active during speech production as an antagonist to
the orbicularis oris inferior muscle (McClean & Sapir, 1980). In addition, the platysma is thought
to be active in individuals during swallow, although this activation is not highly correlated with
27
overall muscle activation during swallow and varies widely between individuals (Palmer,
Luschei, Jaffe, & McCulloch, 1999). Although it is an extremely thin sheath of muscle,
whenever active the platysma will be a substantial source of activity detected at the neck surface
due to its relatively superficial location compared to surrounding muscles.
Resistance against manual force elicits strong consistent activation of strap musculature for
purposes of confirmation of electrode placement and MVC recordings (e.g., Stepp et al., 2011b).
Resistance against manual force can be achieved by mounting an athletic chin guard or similar
apparatus to a dynamometer, to allow for simultaneous collection of force data. Collection of
force data in concert with sEMG during MVC maneuvers can help investigators to improve
reliability of multiday recordings.
Muscles of Respiration
Early work to study the muscles of respiration using sEMG examined the internal and external
intercostals during forced respiration and simple speech tasks (Jones, Beargie, & Pauley, 1953).
Eblen reviewed the subject of using sEMG to record speech-related respiratory muscle activity in
1963, and suggested that sEMG was of limited use due to the inability to isolate activity of
individual muscles (Eblen, 1963). However, McFarland & Smith carefully assessed the ability to
use sEMG to study primary and accessory respiratory musculature during speech and non-speech
tasks (McFarland & Smith, 1989), finding that sEMG could be used to record respiratory-related
activations from the rib cage during speech, but particularly during expiration. They placed
electrodes in a bipolar configuration on the medial and lateral rib cage. The medial set of
electrodes was placed on the 7
th
and 8
th
interspaces, 2 cm lateral to the midclavicular line with
the goal of recording from the diaphragm and intercostals. The lateral set of electrodes was
placed on the 8
th
interspace, straddling the anterior axillary line, also with the goal of detecting
28
activity of the diaphragm and intercostals. Although expiratory activations could be consistently
detected from the rib cage, they were not able to consistently record from the diaphragm or other
muscles of inspiration during speech using electrodes placed on the chest wall. However,
inspiratory related activity was measured from the chest wall when subjects were at higher lung
volumes than those used during typical conversational speech.
Although sEMG can be used to record from abdominal muscles during respiration and speech
(McFarland & Smith, 1989), this area is prone to significant levels of subdermal fat and thus
degraded signal quality. McFarland and Smith recorded from the rectus abdominis by placing
one electrode 2 cm from the midline and 3 cm superior to the umbilicus, with the second
electrode 3 cm directly superior. They sampled activity from the internal and external oblique by
placing electrodes over the anterior axillary line midway between the anterior iliac crest and
costal margin, again using an electrode separation of 3cm (McFarland & Smith, 1989).
Future Directions
Most of the research in sEMG has been accomplished in limb musculature. However, there is
evidence to suggest that there are multiple differences in the anatomy and physiology between
these muscles and the muscles of speech and swallowing. One example is the difference in
discharge rate. Both genioglossus and sternohyoid musculature have been shown to have higher
and more variable discharge rates than limb musculature (Bailey, Rice, & Fuglevand, 2007;
Farina & Falla, 2009). Speech musculature is not load-bearing and differs in many ways from the
more studied muscles of the upper and lower limbs. Additional research is needed to understand
these differences and to more specifically characterize muscle electrophysiology of the speech
and swallowing systems.
29
Before sEMG may be dependably applied clinically, evidence is needed for the reliability of
repeated sEMG measurements. Although issues of reliability of sEMG have been investigated in
other systems, there is still much work to do for speech anatomy to understand how
methodological factors such as electrode placement affect the accuracy and reliability of the
detected activity. Studies in the upper limb have indicated that the most prominent effect on
intersession reliability of sEMG is electrode placement (Yang & Winter, 1983). Thus,
standardized recommendations for electrode placement and orientation will increase the potential
for reliable sEMG recordings of speech anatomy. Digital photography of electrode placement
and careful re-application using anatomical landmarks may also aid in the intersession reliability
of these recordings. Use of even the highest quality commercial recording systems will result in
inaccurate or unreliable data if sensors are improperly or inconsistently placed.
Current technology in common use does not allow isolated muscle recordings of speech
musculature using sEMG. However, future research into flexible high-density arrays of
electrodes may allow for noninvasive characterization of the behavior of isolated muscles. High-
density sEMG in which 2D arrays of electrodes are utilized are becoming increasingly common
in research (Merletti et al., 2009). These arrays consist of several closely spaced small electrodes,
and can be used to estimate the locations of innervation zones and muscle fiber orientations (e.g.,
Lapatki et al., 2010) and in concert with new signal processing techniques to decompose the
sEMG signal into multiple single MUAP trains (e.g., Nawab, Chang, & De Luca, 2010). High-
density electrodes could be of particular use to speech researchers, providing a potential solution
to small, overlapping musculature in which innervation zones are unknown. Although not yet
widely used, researchers have developed flexible high-density electrode grids, which have
30
already been shown to be well-suited to facial recording applications (Lapatki, Van Dijk, Jonas,
Zwarts, & Stegeman, 2004).
Despite the challenges associated with recording from speech and swallowing musculature, the
currently available sEMG still has potential as a clinical and research tool when used correctly.
In a study examining the viability of sEMG recordings of lip muscles, Blair and Smith argue that
even when isolated muscles cannot be recorded, sEMG can be successfully employed for
between condition or group comparisons when consistent electrode placements are used (Blair &
Smith, 1986). Even with its limitations, sEMG can be used to great effect to provide assistance
and rehabilitation for individuals with disordered speech. For example, even without an attempt
to isolate individual muscle groups, multi-channel sEMG can provide successful recognition of
mouthed speech (subvocal) in both control speakers and individuals with dysarthia (Deng et al.,
2009; Meltzner et al., 2008). If reviewers and consumers of sEMG literature demand high quality
methodology and reporting, we will begin to achieve the full potential of this tool and truly affect
patient care through improved assessment and rehabilitation using sEMG.
Acknowledgement
Thanks to Dr. Eric Larson, Deanna Britton, and Mark Malhotra for helpful comments.
31
References
Allen, K. D., Bernstein, B., & Chait, D. H. (1991). EMG biofeedback treatment of pediatric
hyperfunctional dysphonia. Journal of Behavior Therapy and Experimental Psychiatry,
22(2), 97-101.
Allison, G. T., Marshall, R. N., & Singer, K. P. (1993). EMG signal amplitude normalization
technique in stretch-shortening cycle movements. Journal of Electromyography &
Kinesiology, 3(4), 236-244.
Andreassen, S., & Rosenfalck, A. (1978). Recording from a single motor unit during strong
effort. IEEE Transactions on Biomedical Engineering, 25(6), 501-508.
Andrews, S., Warner, J., & Stewart, R. (1986). EMG biofeedback and relaxation in the treatment
of hyperfunctional dysphonia. British Journal of Disorders of Communication, 21(3),
353-369.
Bailey, E. F., Rice, A. D., & Fuglevand, A. J. (2007). Firing patterns of human genioglossus
motor units during voluntary tongue movement. Journal of Neurophysiology, 97(1), 933-
936.
Basmajian, J. V. (1978). Muscles alive: Their functions revealed by electromyography (4th ed.).
Baltimore, MD: The Williams & Wilkins Company.
Beck, T. W., Housh, T. J., Cramer, J. T., Stout, J. R., Ryan, E. D., Herda, T. J., . . . Defreitas, J.
M. (2009). Electrode placement over the innervation zone affects the low-, not the high-
frequency portion of the EMG frequency spectrum. Journal of Electromyography &
Kinesiology, 19(4), 660-666.
Blair, C., & Smith, A. (1986). EMG recording in human lip muscles: Can single muscles be
isolated? Journal of Speech and Hearing Research, 29(2), 256-266.
32
Boucher, V. J., Ahmarani, C., & Ayad, T. (2006). Physiologic features of vocal fatigue:
Electromyographic spectral-compression in laryngeal muscles. Laryngoscope, 116(6),
959-965.
Brown, P., Farmer, S. F., Halliday, D. M., Marsden, J., & Rosenberg, J. R. (1999). Coherent
cortical and muscle discharge in cortical myoclonus. Brain, 122 ( Pt 3), 461-472.
Burnett, T. A., Mann, E. A., Cornell, S. A., & Ludlow, C. L. (2003). Laryngeal elevation
achieved by neuromuscular stimulation at rest. Journal of Applied Physiology, 94(1),
128-134.
Castroflorio, T., Farina, D., Bottin, A., Debernardi, C., Bracco, P., Merletti, R., . . . Bramanti, P.
(2005). Non-invasive assessment of motor unit anatomy in jaw-elevator muscles. Journal
of Oral Rehabilitation, 32(10), 708-713.
Clancy, E. A., & Hogan, N. (1999). Probability density of the surface electromyogram and its
relation to amplitude detectors. IEEE Transactions on Biomedical Engineering, 46(6),
730-739.
Cooper, F. S. (1965). Research techniques and instrumentation: EMG. Proceedings of the
Conference: Communicative Problems in Cleft Palate, ASHA Reports, No. 1, 153-168.
Crary, M. A., & Groher, M. E. (2000). Basic concepts of surface electromyographic biofeedback
in the treatment of dysphagia: A tutorial. American Journal of Speech-Language
Pathology, 9(2), 116-125.
De la Barrera, E. J., & Milner, T. E. (1994). The effects of skinfold thickness on the selectivity of
surface EMG. Electroencephalography and Clinial Neurophysiology, 93(2), 91-99.
De Luca, C. J. (1997). The use of surface electromyography in biomechanics. Journal of Applied
Biomechanics, 13(2), 135-163.
33
De Luca, C. J., Gilmore, L. D., Kuznetsov, M., & Roy, S. H. (2010). Filtering the surface EMG
signal: Movement artifact and baseline noise contamination. Journal of Biomechanics,
43(8), 1573-1579.
Deng, Y., Patel, R., Heaton, J. T., Colby, G., Gilmore, L. D., Cabrera, J., . . . Meltzner, G. S.
(2009). Disordered speech recognition using acoustic and sEMG signals. Paper
presented at the 10th Annual Conference of the International Speech Communication
Association (INTERSPEECH), Brighton, United Kingdom.
Denny, M., & Smith, A. (1992). Gradations in a pattern of neuromuscular activity associated
with stuttering. Journal of Speech and Hearing Research, 35(6), 1216-1229.
Denny, M., & Smith, A. (2000). Respiratory control in stuttering speakers: Evidence from
respiratory high-frequency oscillations. Journal of Speech, Language, and Hearing
Research, 43(4), 1024-1037.
Disselhorst-Klug, C., Schmitz-Rode, T., & Rau, G. (2009). Surface electromyography and
muscle force: Limits in sEMG-force relationship and new approaches for applications.
Clinical Biomechanics, 24(3), 225-235.
Eblen, R. E. (1963). Limitations on use of surface electromyography in studies of speech
breathing. Journal of Speech and Hearing Research, 6(1), 3-18.
Falla, D., Dall'Alba, P., Rainoldi, A., Merletti, R., & Jull, G. (2002). Location of innervation
zones of sternocleidomastoid and scalene muscles--a basis for clinical and research
electromyography applications. Clinical Neurophysiology, 113(1), 57-63.
Farina, D., Cescon, C., & Merletti, R. (2002). Influence of anatomical, physical, and detection-
system parameters on surface EMG. Biological Cybernetics, 86(6), 445-456.
34
Farina, D., & Falla, D. (2009). Discharge rate of sternohyoid motor units activated with surface
EMG feedback. Journal of Neurophysiology, 101(2), 624-632.
Farina, D., Merletti, R., & Disselhorst-Klug, C. (2004). Multi-channel techniques for information
extraction from the surface EMG. In R. Merletti & P. Parker (Eds.), Electromyography:
Physiology, engineering, and non-invasive applications (pp. 169-203). Hoboken, NJ:
John Wiley & Sons, Inc.
Farina, D., Merletti, R., & Stegeman, D. F. (2004). Biophysics of the generation of EMG signals.
In R. Merletti & P. A. Parker (Eds.), Electromyography: Physiology, engineering, and
non-invasive applications (pp. 81-105). Hoboken, NJ: John Wiley & Sons, Inc.
Farina, D., & Rainoldi, A. (1999). Compensation of the effect of sub-cutaneous tissue layers on
surface EMG: a simulation study. Medical Engineering and Physics, 21(6-7), 487-497.
Fuglevand, A. J., Winter, D. A., Patla, A. E., & Stashuk, D. (1992). Detection of motor unit
action potentials with surface electrodes: influence of electrode size and spacing.
Biological Cybernetics, 67(2), 143-153.
Gay, T., & Harris, K. S. (1971). Some recent developments in the use of electromyography in
speech research. Journal of Speech and Hearing Research, 14(2), 241-246.
Goffman, L., & Smith, A. (1994). Motor unit territories in the human perioral musculature.
Journal of Speech and Hearing Research, 37(5), 975-984.
Grosse, P., Cassidy, M. J., & Brown, P. (2002). EEG-EMG, MEG-EMG and EMG-EMG
frequency analysis: physiological principles and clinical applications. Clinical
Neurophysiology, 113(10), 1523-1531.
Halliday, D. M., Rosenberg, J. R., Amjad, A. M., Breeze, P., Conway, B. A., & Farmer, S. F.
(1995). A framework for the analysis of mixed time series/point process data--theory and
35
application to the study of physiological tremor, single motor unit discharges and
electromyograms. Progress in Biophysics and Molecular Biology, 64(2-3), 237-278.
Hermens, H. J., Freriks, B., Disselhorst-Klug, C., & Rau, G. (2000). Development of
recommendations for SEMG sensors and sensor placement procedures. Journal of
Electromyography & Kinesiology, 10(5), 361-374.
Hermens, H. J., Freriks, B., Merletti, R., Stegeman, D. F., Blok, J. H., Gunter, R., . . . Hagg, G.
M. (1999). European recommendations for surface electromyography: Results of the
SENIAM project. Enschede: Roessingh Research and Development.
Hocevar-Boltezar, I., Janko, M., & Zargi, M. (1998). Role of surface EMG in diagnostics and
treatment of muscle tension dysphonia. Acta Otolaryngologica, 118(5), 739-743.
Huckabee, M. L., & Cannito, M. P. (1999). Outcomes of swallowing rehabilitation in chronic
brainstem dysphagia: A retrospective evaluation. Dysphagia, 14(2), 93-109.
Jones, D. S., Beargie, R. J., & Pauley, J. E. (1953). An electromyographic study of some muscles
on costal respiration in man. The Anatomical Record, 117(1), 17-24.
Kamen, G., & Caldwell, G. E. (1996). Physiology and interpretation of the electromyogram.
Journal of Clinical Neurophysiology, 13(5), 366-384.
Koh, T. J., & Grabiner, M. D. (1992). Cross talk in surface electromyograms of human
hamstring muscles. Journal of Orthopaedic Research, 10(5), 701-709.
Koh, T. J., & Grabiner, M. D. (1993). Evaluation of methods to minimize cross talk in surface
electromyography. Journal of Biomechanics, 26 Suppl 1, 151-157.
Koole, P., de Jongh, H. J., & Boering, G. (1991). A comparative study of electromyograms of the
masseter, temporalis, and anterior digastric muscles obtained by surface and
intramuscular electrodes: raw-EMG. Cranio, 9(3), 228-240.
36
Kuiken, T. A., Lowery, M. M., & Stoykov, N. S. (2003). The effect of subcutaneous fat on
myoelectric signal amplitude and cross-talk. Prosthetics and Orthotics International,
27(1), 48-54.
Lapatki, B. G., Oostenveld, R., Van Dijk, J. P., Jonas, I. E., Zwarts, M. J., & Stegeman, D. F.
(2006). Topographical characteristics of motor units of the lower facial musculature
revealed by means of high-density surface EMG. Journal of Neurophysiology, 95(1),
342-354.
Lapatki, B. G., Oostenveld, R., Van Dijk, J. P., Jonas, I. E., Zwarts, M. J., & Stegeman, D. F.
(2010). Optimal placement of bipolar surface EMG electrodes in the face based on single
motor unit analysis. Psychophysiology, 47(2), 299-314.
Lapatki, B. G., Stegeman, D. F., & Jonas, I. E. (2003). A surface EMG electrode for the
simultaneous observation of multiple facial muscles. Journal of Neuroscience Methods,
123(2), 117-128.
Lapatki, B. G., Van Dijk, J. P., Jonas, I. E., Zwarts, M. J., & Stegeman, D. F. (2004). A thin,
flexible multielectrode grid for high-density surface EMG. Journal of Applied
Physiology, 96(1), 327-336.
Lindstrom, L., Kadefors, R., & Petersen, I. (1977). An electromyographic index for localized
muscle fatigue. Journal of Applied Physiology, 43(4), 750-754.
Loucks, T. M., Poletto, C. J., Saxon, K. G., & Ludlow, C. L. (2005). Laryngeal muscle responses
to mechanical displacement of the thyroid cartilage in humans. Journal of Applied
Physiology, 99(3), 922-930.
37
Maarsingh, E. J., Oud, M., van Eykern, L. A., Hoekstra, M. O., & van Aalderen, W. M. (2006).
Electromyographic monitoring of respiratory muscle activity in dyspneic infants and
toddlers. Respiratory Physiology & Neurobiology, 150(2-3), 191-199.
McClean, M. D., & Sapir, S. (1980). Surface electrode recording of platysma single motor units
during speech. Journal of Phonetics, 8, 169-173.
McClean, M. D., & Tasko, S. M. (2003). Association of orofacial muscle activity and movement
during changes in speech rate and intensity. Journal of Speech, Language, and Hearing
Research, 46(6), 1387-1400.
McFarland, D. H., & Smith, A. (1989). Surface recordings of respiratory muscle activity during
speech: some preliminary findings. Journal of Speech and Hearing Research, 32(3), 657-
667.
Meltzner, G. S., Sroka, J., Heaton, J. T., Gilmore, L. D., Colby, G., Roy, S., . . . De Luca, C. J.
(2008). Speech recognition for vocalized and subvocal modes of production using surface
EMG signals from the neck and face. Paper presented at the 9th Annual Conference of
the International Speech Communication Association (INTERSPEECH), Brisbane,
Australia.
Mercer, J. A., Bezodis, N., DeLion, D., Zachry, T., & Rubley, M. D. (2006). EMG sensor
location: Does it influence the ability to detect differences in muscle contraction
conditions? Journal of Electromyography & Kinesiology, 16(2), 198-204.
Merletti, R., Botter, A., Troiano, A., Merlo, E., & Minetto, M. A. (2009). Technology and
instrumentation for detection and conditioning of the surface electromyographic signal:
state of the art. Clinical Biomechanics, 24(2), 122-134.
38
Merletti, R., & Hermens, H. J. (2004). Detection and conditioning of the surface EMG signal. In
R. Merletti & P. Parker (Eds.), Electromyography: Physiology, engineering and non-
invasive applications (pp. 107-131). Hoboken: Wiley-IEEE Press.
Moritani, T., Stegeman, D. F., & Merletti, R. (2004). Basic physiology and biophysics of EMG
signal generation. In R. Merletti & P. Parker (Eds.), Electromyography: Physiology,
engineering and non-invasive applications (pp. 107-131). Hoboken: Wiley-IEEE Press.
Mu, L., & Sanders, I. (1998). Neuromuscular specializations of the pharyngeal dilator muscles: I.
Compartments of the canine geniohyoid muscle. The Anatomical Record, 250(2), 146-
153.
Nawab, S. H., Chang, S. S., & De Luca, C. J. (2010). High-yield decomposition of surface EMG
signals. Clinical Neurophysiology, 121(10), 1602-1615.
Netto, K. J., & Burnett, A. F. (2006). Reliability of normalisation methods for EMG analysis of
neck muscles. Work, 26(2), 123-130.
Norman, R. W., Nelson, R. C., & Cavanagh, P. R. (1978). Minimum sampling time required to
extract stable information from digitized EMGs. In E. Asmussen & K. Jorgensen (Eds.),
Biomechanics VI-A (Vol. 2A, pp. 237-243). Baltimore, MD: University Park Press.
O'Dwyer, N. J., Quinn, P. T., Guitar, B. E., Andrews, G., & Neilson, P. D. (1981). Procedures
for verification of electrode placement in EMG studies of orofacial and mandibular
muscles. Journal of Speech and Hearing Research, 24(2), 273-288.
Palmer, P. M., Luschei, E. S., Jaffe, D., & McCulloch, T. M. (1999). Contributions of individual
muscles to the submental surface electromyogram during swallowing. Journal of Speech,
Language, and Hearing Research, 42(6), 1378-1391.
39
Pettersen, V., Bjorkoy, K., Torp, H., & Westgaard, R. H. (2005). Neck and shoulder muscle
activity and thorax movement in singing and speaking tasks with variation in vocal
loudness and pitch. Journal of Voice, 19(4), 623-634.
Roeleveld, K., Stegeman, D. F., Vingerhoets, H. M., & Van Oosterom, A. (1997). Motor unit
potential contribution to surface electromyography. Acta Physiologica Scandinavica,
160(2), 175-183.
Roy, S. H., De Luca, G., Cheng, M. S., Johansson, A., Gilmore, L. D., & De Luca, C. J. (2007).
Electro-mechanical stability of surface EMG sensors. Medical and Biological
Engineering and Computing, 45(5), 447-457.
Ruark, J. L., & Moore, C. A. (1997). Coordination of lip muscle activity by 2-year-old children
during speech and nonspeech tasks. Journal of Speech, Language, and Hearing
Research, 40(6), 1373-1385.
Smith, A., & Denny, M. (1990). High-frequency oscillations as indicators of neural control
mechanisms in human respiration, mastication, and speech. Journal of Neurophysiology,
63(4), 745-758.
Stemple, J. C., Weiler, E., Whitehead, W., & Komray, R. (1980). Electromyographic
biofeedback training with patients exhibiting a hyperfunctional voice disorder.
Laryngoscope, 90(3), 471-476.
Stepp, C. E., Heaton, J. T., Braden, M. N., Jetté, M. E., Stadelman-Cohen, T. K., & Hillman, R.
E. (2011a). Comparison of neck tension palpation rating systems with surface
electromyographic and acoustic measures in vocal hyperfunction. Journal of Voice,
25(1), 67-75.
40
Stepp, C. E., Hillman, R. E., & Heaton, J. T. (2010). Use of neck strap muscle intermuscular
coherence as an indicator of vocal hyperfunction. IEEE Transactions on Neural Systems
and Rehabilitation Engineering, 18(3), 329-335.
Stepp, C. E., Hillman, R. E., & Heaton, J. T. (2011b). Modulation of neck intermuscular beta
coherence during voice and speech production. Journal of Speech, Language, and
Hearing Research, 54(3), 836-844.
Suvinen, T. I., Malmberg, J., Forster, C., & Kemppainen, P. (2009). Postural and dynamic
masseter and anterior temporalis muscle EMG repeatability in serial assessments. Journal
of Oral Rehabilitation, 36(11), 814-820.
Tamplin, J., Brazzale, D. J., Pretto, J. J., Ruehland, W. R., Buttifant, M., Brown, D. J., &
Berlowitz, D. J. (2011). Assessment of breathing patterns and respiratory muscle
recruitment during singing and speech in quadriplegia. Archives of Physical Medicine
and Rehabilitation, 92(2), 250-256.
Tassinary, L. G., Cacioppo, J. T., & Geen, T. R. (1989). A psychometric study of surface
electrode placements for facial electromyographic recording: I. The brow and cheek
muscle regions. Psychophysiology, 26(1), 1-16.
van Boxtel, A. (2001). Optimal signal bandwidth for the recording of surface EMG activity of
facial, jaw, oral, and neck muscles. Psychophysiology, 38(1), 22-34.
van Boxtel, A., Goudswaard, P., van der Molen, G. M., & van den Bosch, W. E. (1983). Changes
in electromyogram power spectra of facial and jaw-elevator muscles during fatigue.
Journal of Applied Physiology, 54(1), 51-58.
41
Vigreux, B., Cnockaert, J. C., & Pertuzon, E. (1979). Factors influencing quantified surface
EMGs. European Journal of Applied Physiology and Occupational Physiology, 41(2),
119-129.
Webster, J. G. (1984). Reducing motion artifacts and interference in biopotential recording.
IEEE Transactions on Biomedical Engineering, 31(12), 823-826.
Wentzel, S. E., Konow, N., & German, R. Z. (2010). Regional differences in hyoid muscle
activity and length dynamics during mammalian head shaking. Journal of Experimental
Zoology Part A: Ecological Genetics and Physiology, 315(3), 111-120.
Wohlert, A. B., & Goffman, L. (1994). Human perioral muscle activation patterns. Journal of
Speech and Hearing Research, 37(5), 1032-1040.
Wohlert, A. B., & Hammen, V. L. (2000). Lip muscle activity related to speech rate and
loudness. Journal of Speech, Language, and Hearing Research, 43(5), 1229-1239.
Yang, J. F., & Winter, D. A. (1983). Electromyography reliability in maximal and submaximal
isometric contractions. Archives of Physical Medicine and Rehabilitation, 64(9), 417-
420.
Yiu, E. M., Verdolini, K., & Chow, L. P. (2005). Electromyographic study of motor learning for
a voice production task. Journal of Speech, Language, and Hearing Research, 48(6),
1254-1268.
42
Figure Captions
Figure 1. Example of potential signal aliasing with a sampling frequency less than twice the
signal frequency. The original 200 Hz signal is shown in the solid black line. Sampling at 2,000
Hz (unfilled circles; sampling period of 0.5 ms) would provide information necessary to
reconstruct the original signal. Sampling at 267 Hz (filled circles; sampling period of 3.75 ms)
results in a reconstructed signal (dashed grey line) that is equivalent to a 67 Hz signal. Thus,
undersampling can result in aliasing of signal energy from the original signal frequency to a
lower frequency.
Figure 2. sEMG recordings illustrating high signal to noise ratio (A) and poor signal to noise
ratio (B) with contamination from 60 cycle noise. Panel A: Anterior neck sEMG from a young
healthy participant with minimal subdermal neck fat during reading aloud. Left plots show
sEMG as a function of time. Right plot shows the power spectral density the sEMG using
Welch's method. Panel B: Anterior neck sEMG from a middle-aged participant with significant
levels of subdermal fat, leading to a poor signal-to-noise ratio and contamination of the signal
with electrical line noise. Electrical line noise is apparent in both the power spectral density (as
peaks of energy energy at 60 Hz and its harmonics) as well as the time trace of the sEMG at
close time intervals.
Figure 3. Diagram of neck muscles as seen from the front illustrating examples of bipolar
electrode configurations. Panel A: Incorrect bilateral configuration. Panel B: Suggested
configuration with electrodes placed parallel to the longitudinal axis of the muscle body, in line
with the fibers of the muscle. TH = thyrohyoid, OH = omohyoid, SCM = sternocleidomastoid,
SH = sternohyoid, ST = sternothyroid.
43
Tables
Table 1. Gestures known to elicit strong (not isolated) activation from each muscle (O'Dwyer et
al., 1981).
Muscle Gesture
Levator Labii Superioris Unilateral snarl elevating the
upper lip
Zygomaticus Major Broad laugh
Buccinator Puffing out the cheeks with
the lips closed
Risorius Broad smile with the lips
closed
Orbicularis Oris Superioris Compressing the upper lip
against the upper incisors
Orbicularis Oris Inferioris Compressing the lower lip
against the lower incisors
Depressor Anguli Oris Pulling down the corners of
the mouth
Depressor Labii Inferioris Pulling down the lower lip
with the jaw closed
Mentalis Raising and everting the
lower lip while wrinkling the
chin
Mylohyoid Swallowing
Anterior Belly of the
Digastric
Lowering jaw under
resistance
