.
.
.
.
.
.
.
There is a well-established relation between the size of the open velar port and the position
of the velum. In addition though to varying with port area, velar position also varies when the
port is completely closed. These adjustments result from the anatomical relationship between
the velum and the levator palatini muscle. Since the muscle's superior attachment lies
well above the level at port closure is complete, increasing contraction of this muscle continues
to raise the velum even after closure has occurred. As a result, changes in the vertical position of
the velum throughout its range of movement are presumed to reflect general velar control
strategies and our study of them is not limited to pieces of speech having an open port. Being
able to monitor the vertical position of the velum throughout its range should aid in the
discovery of the principles of normal velar motor control as well as increasing our ability to
evaluate velar control problems in some clinical populations. The velar trace was developed to
allow us to monitor these changes in velar position. It allows us to collect mid-sagittal velar position
data in analog form, eliminating x-ray exposure to the subject and frame-by-frame measurements
of cine and video recordings by the investigator. In this drawing of the velo trace we see its three
major parts, an internal lever, an external lever, and a push rod between them, all carried on a
thin support rod. The levers are connected to the push rod so that when the internal lever is raised,
the external lever is deflected toward the subject. The device is loaded with a small spring to improve
its frequency response and thus the ability of the internal lever to follow rapid downward
movements of the velum. The external lever can be monitored in several different ways. For example,
one might use a velocity displacement transducer that would make the velo trace a convenient
standalone device for the clinical evaluation of velar movement. This diagram shows another approach,
optoelectronic tracking. In the system we've been using, a light emitting diode or LED is attached
to the end of the external lever to allow us to monitor the movement of the lever about its fulcrum.
A second LED is positioned at the end of the support rod and a third near the bridge of the
nose. These serve as reference points against which the movements of the external lever can
be described, including making corrections for head movement. The positions of the LEDs are tracked in
two-dimensional space and recorded simultaneously with the acoustic signal and a timing signal on a
multi-channel instrumentation data recorder. Before inserting the velo trace, topical anesthetic and
decongestants are applied to the nasal mucosa. The velo trace is then inserted using a procedure
similar to that used for nasal catheterization and although it's a rigid device, the insertion is
easy unless the subject has serious nasal pathologies or deformities.
Finally, an external clamp attached to a headband is used to stabilize the position of the velo
trace over the course of the recording session. Here we see the relative size and positioning of
the velo trace in a life-size model. Several studies were done to evaluate the velo trace.
In one study, we examined the ability of the velo trace to follow rapid downward and upward
movements of the velum. Velo trace data were compared with data obtained through frame-by-frame
measurements of cine films photographed through a nasally positioned fiber optic endoscope and
some of those data are shown in this figure. Veloposition time functions are shown for two
different test words. FESMEAP in the left panels and PHEMZEEP in the right panels. Data were
recorded with an endoscope for one subject, the upper panels, and with the velo trace for another
subject. In all cases, the traces represent ensemble averages of a set of tokens of the
same utterance type. Clearly, the patterns are similar for the ensemble averaged velo trace
data and the ensemble averaged frame-by-frame endoscope data despite the fact that the data
were obtained from two different speakers. In a second evaluation study, Cineradiographic
films and velo trace data were obtained simultaneously from a single subject. The
velo trace was positioned in one of the subject's nasal cavities and a thin gold chain was inserted
through the other nasal cavity and positioned along the vellum and down into the oropharynx.
The frame-by-frame measurements made of the highest point of the vellum and of the vertical
position of the tip of the internal lever are seen in this figure. The very high linear
correlation between these measures is reflected in this scatter plot and the correlation coefficient
of 0.993. In addition, velocity measures obtained from our velo trace data were consistent with
velo velocity data reported in the literature. It appears that movements of the velo trace reflect
the movements of the vellum itself. Here is a subject producing repetitions of phrases used
in a recent study. It's a lonsal again. It's a lonsal again. It's a lonsal again. It's a lonsal
again. It's a lonsal again. It's a lonsal again. It's a lonsal again. It's a lonsal again. It's a
lonsal again. It's a lonsal again. These traces represent movements of the external lever for the
utterance we just heard. It's salasal again. The vellum moves up for the ts of its, then down a bit
for a, l, and a, then up again for s, and then g, and down for the n of again. Now let's look at some
data. This figure shows velocity traces, movement patterns, and acoustic waveforms for three
utterances. It's lonsal again. It's a lonsal again. And it's salonsal again. The data were corrected
for head movement, and the signal was smoothed. We marked the beginning of the nasal consonant n
in the acoustic waveform, and also the beginning of each vellum lowering movement between the s of
its and the n of lonsal, using the velocity trace to determine the beginning of each movement. In
each case, the large lowering movement for n is highlighted in yellow. In the longer sequences,
we also found an early shallow movement that we've highlighted in purple. This figure shows how
revealing comparisons between oral and nasal sequences can be. These are movement traces for
sequences with and without a nasal consonant. In the top panel, you see the sequences. It's a
lonsal again, and it's a lonsal again, superimposed on each other. Notice that the vellum moves from
its high position for the s of its into a lower position in both kinds of sequence. However,
in the oral sequence, the vellum then rises for the s of lonsal, whereas in the nasal sequence,
it continues lowering for the nasal consonant of lonsal. This figure shows that the early shallow
movement in the nasal string is related to the string a la, while the later more extreme movement
is associated with the nasal consonant. The lower figure shows the corresponding data for its say la
sal again and its say lonsal again. Here, we see again that the oral sequence contains lowering
movements that are mirrored in the nasal sequence, and that the lowering for the nasal consonant
itself occurs relatively late. We use these data to help us understand the interaction between
intrinsic segmental requirements of the vellum and extent of velar coarticulation for a nasal
consonant. Here, we presented them to show you how this device works, what kind of data can be
obtained, and how those data can be analyzed. Obtaining the acoustic signal simultaneously with
the velotrace signal made it possible to identify the acoustic segments and processing the
velotrace signal to obtain the instantaneous velocity made it possible to identify the
beginning and end of each movement. Although we did not obtain movement data for other
articulators in this study, we have in other studies collected lip and jaw movement data
along with the velotrace data. In those, we examined interarticulator coordination in the
production of nasal consonants. We have found that velotrace data are relatively easy to obtain and
allow us to use efficient analysis procedures while avoiding the risk of x-ray exposure for the subject.
The details of tongue movements during speech are of great interest to anyone involved in
understanding the processes of speech communication, particularly phoneticians
and speech and language pathologists. With electropalletography, or EPG for short,
it's possible to record one important aspect of tongue activity, the location and timing of tongue
contacts with a hard palate during connected speech. First, we construct a thin acrylic
palate made from a plaster impression. It has 62 silver electrodes serving as contact points
embedded in the material. The rows of electrodes are placed according to anatomical landmarks.
For example, the back row is at the junction between the hard and soft palates and the front
row at the place where the upper front teeth meet the alveolar ridge. Rows are placed proportionately
on the surface, with the front four rows being closer together than the back four rows. Wires
from each electrode are fed into two cables which pass out of the subject's mouth at the corners,
usually behind the wisdom teeth, so as to interfere as little as possible with normal speech. Once the
subject has the palate in place, she plugs into a multiplexer unit worn around her neck and connects
to the computer interface. A circuit is completed by means of a hand-held electrode and contact
patterns are now displayed on the computer screen. The points on the screen represent the electrode
positions with the alveolar region at the top. Contact is shown by filled white squares. Note
that the computer display shows a symmetrical arrangement of the electrodes. In actual fact,
the front four rows, as seen earlier, are concentrated in the anterior parts of the palate.
The rows of electrodes can be related to the place of articulation labels commonly used in phonetics.
We will now demonstrate the contact patterns of some English sounds. EPG is particularly useful
for showing place of articulation for alveolar and velar plosives. First, alveolar plosives.
Ah ta. Ah ta. During the closure phase of the plosive, note the full contact at the alveolar
ridge and along both sides. Next, velar plosives. Ah ta. Ah ta. Full closure is now at the velar region.
Coarticulatory effects can be shown very clearly with EPG. Compare, for example,
Ah ca. Ah quai. Ah ca. Ah quai. Ah ca. Ah quai.
Note the greater amount of contact for the plosive in ah quai, reflecting the higher and more forward
placement of the tongue in anticipation of the e vowel. Different manners of articulation can
also be illustrated with EPG. Alveolar and post-alveofricative show characteristic patterns
of contact. First, alveolar fricatives. Ah sa. Ah sa. A narrow central groove can be seen clearly
along with full lateral contact. Next, post-alveolar fricatives. Ah sha. Ah sha. As with the alveolar
fricative, a groove is present but it's wider and in a more retracted position. Lateral approximants
also show distinctive patterns of contact which depend greatly on the vowel environment. Compare,
for example, Ah la. Ah la. And Ah le. Ah le. The dynamics of tongue activity in more exotic sounds
can be studied with EPG. For example, Ah ra. Ah ra. A slow motion replay reveals the individual
alveolar contacts during production of the trill. Sounds produced with a velaric airstream mechanism
such as clicks reveal interesting information about tongue dynamics. Ah sa. Ah sa. An alveolar
click shows near simultaneous alveolar and velar contact. The technique can thus provide
very revealing information on the dynamics of tongue activity for specific sounds. With
the computer, it's possible to examine what happens to these sounds in connected speech.
I recorded a sentence here and we can analyse the EPG signal along with the acoustic waveform.
First let's hear the sentence. As I move the cursor through the sentence, we can see changing
patterns of EPG contact. I'm placing the cursor here at the end of the word KED where you
would expect a D with alveolar contact. Note, however, that the D is completely assimilated
into the place of articulation of the following velar closure. Moving further into the utterance,
at the end of the first vowel in the word cocktail, note first the velar closure, then
the alveolar with a brief period of simultaneous or double articulation. This display shows
a sequence of palatal frames sampled at 10 millisecond intervals for the initial plosive
in the word TED. From this full contact data, a graph can be constructed showing the total
number of contacts occurring in different regions of the palate. Here we see a display
showing the total number of contacts in three regions of the palate for the first hundred
frames in the utterance. Another data reduction facility is a centre of gravity plot. Centre
of gravity shows a concentration of contacts in different parts of the palate. Here the
centre of gravity is plotted as a function of the total number of electrodes. The graph
shows movement in the centre of gravity from the front of the palate to the right of the
graph to the back in the word TED, spoken here with the assimilated final plosive. This
plot shows a frequency of contact of each electrode over a specified portion of the
utterance. EPG is thus very suited for studying the sorts of connected speech processes that
occur when we speak normally and spontaneously. EPG is proving to be a valuable clinical tool
for assessing and treating a wide range of speech disorders. The technique can provide
details of tongue movements that aren't always obvious from an auditory analysis. Peter was
aged nine and a half years when he was referred for EPG therapy. He was born with a unilateral
cleft lip and palate. This was repaired surgically before he was a year old. Despite early surgery,
his speech problems persisted through the preschool years and this might have been hindered
by conductive hearing loss. Recent video fluoroscopy had shown his velopharyngeal competence to
be normal. This is Peter's production of a biscuit. Production of the grooved sibilant
S was obviously distorted with nasal escape of air. The EPG patterns during target S show
complete constriction in the velar region. Interestingly, this feature of complete contact
in the velar region occurred not just for S but for all anterior obstruents. The cursor
is now placed during the bilabial stop in the word biscuit. In a normal speaker, there
would be minimal tongue palate contact here but Peter shows evidence of complete velar
closure. The cursor is now placed during the final alveolar closure in the word biscuit.
Note the abnormal double alveolar velar articulation. These double articulations were not obvious
from the auditory analysis. Subsequent therapy for Peter involved inhibiting these abnormal
double articulations as well as trying to establish a more normal looking alveolar groove
configuration for S. The type of visual feedback provided by EPG helps children to understand
exactly which articulatory gestures they're producing incorrectly. It also gives them
immediate feedback when they try to monitor or change their own articulatory patterns.
In a recent project funded by the British Medical Research Council, we used EPG to record
and treat 20 school aged children with articulation disorders which had proved to be resistant
to conventional therapy. The results so far have proved very positive. Here are some examples
of EPG patterns before and after therapy. These EPG patterns are from five children
who took part in the project. The target sound is the S in a castle. One frame only has been
selected, that of maximum constriction during S production. The five children had a variety
of abnormal patterns for S before therapy. After therapy with EPG, the patterns more
closely resembled a normal S configuration and they were also heard as acceptable productions.
In the future, we need to continue to construct and carry out clinical trials using EPG so
that clinicians can begin to recognise which client groups can benefit from this new approach
to therapy. So EPG offers exciting new prospects both as a laboratory and a clinical technique.
Thank you.
Since speech production is very variable, large quantities of kinematic data are needed
to help uncover underlying principles of speech motor control.
Therefore, we should have efficient techniques to gather and analyze articulatory data.
Genetic midsagittal articulometer systems, which we call EMMA systems, make it possible
to record movements of multiple midline points on vocal tract structures.
Although these systems can be expensive, they are within reach of individual laboratories.
If used correctly, they can provide the needed quantities of accurate articulatory data with
minimal risk to experimental subjects.
In the EMMA system that we developed and use at MIT, three transmitter coils are held in
a transmitter assembly, with the coil axes perpendicular to the midline plane.
The transmitter assembly is positioned on a subject, so its midline coincides with the
midsagittal plane.
Each transmitter is excited by a sinusoidal signal at a different frequency, between 60
and 80 kilohertz, generating an alternating magnetic field with a strength that decreases
in proportion to the cube of the distance from the transmitter.
Small encased transducer coils are mounted on the subject's articulators.
They can be mounted on the tongue blade, tongue body, lower incisors, lips, and possibly the
velum.
Transducers are also mounted on the bridge of the nose and upper central incisors for
a maxillary frame of reference.
In order to obtain accurate data, the transducers are mounted as close as possible to the midline,
with their axes parallel to the transmitter axes.
The alternating magnetic fields from the transmitters induce alternating voltages in the transducers,
which are connected to receiver electronics with fine insulated wires.
The electronics convert the induced high frequency signals to three slowly varying output signals
from each transducer.
These signals are digitized along with the speech signal.
We can also digitize simultaneously the output of other transducers.
The digitized signals are converted by signal processing software to X and Y coordinates
in the midline plane.
As the articulators move, the transducer axes can vary in their alignment with the transmitter
axes, causing potential measurement error.
The signal processing software includes a calculation that corrects for this rotational
misalignment.
The field strengths and frequencies delivered to a subject by our EMA system are similar
to the ones experienced by the user of an engineering workstation.
Generally, it's thought that such fields are not harmful to non-pregnant adults.
But since there are unanswered questions about this issue, we pay careful attention to new
studies of the effects of magnetic fields like these on humans.
Before each experimental run, we perform two types of calibrations, both of which are needed
by the signal processing to produce accurate measurements.
We calibrate the fields to obtain exponents that characterize how the field strengths
change with distance from the transmitters.
Data are generated for the field calibrations by moving a transducer through several circles
of known radius.
The transducers are calibrated to obtain individual gain and offset scale factors.
We gather data for each transducer calibration by placing it in a number of known locations
in the midline plane.
The subject is seated in a comfortable position, and the transmitter assembly is placed over
and secured to his head.
The assembly is positioned vertically and horizontally, so the center of the measurement
plane coincides with the subject's oral cavity.
We check and adjust the alignment of the assembly midline with the subject's midsagittal plane,
using the midline pointer being manipulated by the experimenter.
Then we put midline marks on the subject's face as a guide to transducer placement.
Next, we mount the transducers on the subject's articulators.
We use a biocompatible rubber cement for extra oral placements and a biocompatible cyanoacrylate
for intraoral placements.
As we mount each transducer, we position it as close as possible to the midline, and we
keep its axis perpendicular to the midsagittal plane.
In addition to accurate calibrations, the integrity of the data depends very much on
the careful placement of transducers on the subject.
We check each placement to be sure that no transducer is more than 3 millimeters from
the midline, because the misalignment correction algorithm is increasingly less effective,
with transducer placements that are further from the midline.
This time, in attempting to provide a clearer view for the video camera, we placed some
transducers unusually far to the subject's right, very near the tolerable limit.
Normally, we would reposition these transducers.
We can record about an hour's worth of data.
The recording time is usually limited by how long the intraoral transducers remain mounted
securely.
While the subject reads the text, an experimenter checks to be sure that the transducers remain
in place, and that the subject continues to be aligned in the transmitter assembly.
At the same time, the transduced positions are monitored on a real-time display to check
for changes that might indicate loosening of an attachment or excessive misalignment.
The red hood hit it, the red hood hit it.
We also make a video recording of the subject's lips.
This recording is synchronized with the EMMA recording, so we can relate EMMA data on lip
positions to labial cross-sectional areas.
When the speech data have been collected, we record the shape of the subject's palate
by drawing a transducer along its midline.
In addition, we record the orientation of the occlusal plane by having the subject bite
lightly on a plate that has two transducers mounted on it.
The more anterior transducer is positioned just forward of the upper incisors.
When the experiment is over, the transducers are removed carefully from the subject, and
the transmitter assembly is lifted off.
Next, we run a program that performs a signal processing.
For each transducer, the resulting signals consist of x and y components of displacement,
velocity and acceleration versus time, as well as time-varying magnitudes of velocity
and acceleration, and a measure of the amount of misalignment correction called the correction
index.
The transducer coordinates are calculated with respect to the fixed maxillary transducer
locations, and they are translated and rotated to the occlusal plane frame of reference.
We examine the results with a program that allows us to listen to the speech and label
events on the waveforms.
In this display, the top window shows the RMS of the speech signal for five utterances.
We can listen to the signal to locate the desired utterance.
Ma who hit it, ma who had it, ma who hit it, ma who hit it, ma who had it.
Now we delimit the utterance we are interested in with two cursors, which causes the delimited
speech signal and a selected set of articulatory signals to be displayed below on the left.
For orientation purposes, y versus x trajectories are shown for transducers on the tongue body,
lower incisor, lower lip, and upper lip, along with an outline of the hard palate.
The articulatory signals at the left are the values of the misalignment correction index
versus time for the four transducers.
As is typical, the tongue transducer shows the largest amount of index variability.
We check the correction index for abnormally large values, and we listen to the pronunciation
of the utterance.
If these are acceptable, we enter a phonetic label to identify the utterance.
If not, we bypass the token.
Here, the signals on the lower left are for the tongue body x and y coordinates and the
absolute value of tongue body velocity versus time for the utterance ma who hit it.
At the lower right is an expanded plot of tongue body y versus x.
The spacing between the points corresponds to the 3.2 millisecond sampling period.
The scale bars show one millimeter.
The two channel numbers mark the beginning of the trajectory.
As we place the cursor on the time signals, a crosshair is displayed on the trajectory
at the selected time.
Here we can see the locations on the trajectory that correspond to the time of the beginning
of the ah, the end of the ah, the beginning of the ooh, and the end of the ooh.
At the time near the middle of the ooh, there is a velocity minimum which corresponds to
the extreme extent of tongue body movement towards the ooh target.
For subsequent analysis, we label the beginning and end of the ooh with the help of the display
of the expanded acoustic signal shown above the trajectory plot.
Next, the labeled acoustic events are used to guide algorithmic data extraction.
The signals on the lower left are velocity magnitude versus time for the four transducers.
On the lower right are expanded xy trajectories.
First, mid-valve spectra of the acoustic signal are displayed with algorithmically picked
LPC peaks.
Then the cursor and crosshairs move automatically to the time of the minimum velocity for each
transducer for extraction of x and y coordinates.
At each step, the experimenter checks the operation of the algorithm.
With experimenter approval or correction if necessary, formant values, transducer coordinates,
minimum velocity times, and correction index values are written to a text file for subsequent
graphical and statistical analysis.
This system is flexible so it can be programmed readily to perform a wide variety of data
extraction algorithms.
It can also be helpful to examine articulatory trajectories for multiple repetitions of an
utterance to investigate directions of movement, trajectory shapes, and token-to-token variability.
In this example, we see a y versus x trajectory for the tongue body in pronunciation of the
utterance he ga.
The numbers on the trajectories indicate labeled times of acoustic events.
They correspond to the onset of the e, onset of the g, release of the g, onset of the a,
and end of the a.
The g is produced with a forward tongue body movement which, because of the context, results
in a looping trajectory.
Examination of a sequence of such plots for multiple utterances shows some variation but
a rather consistent looping trajectory for the g.
It can also be helpful to show a number of such trajectories on a single display for
comparison purposes.
With these systems, there is no way to determine absolutely how large the measurement error
might be, so it's important to explore the data for indications of possible measurement
error.
To do so, we examine time series plots of extracted data.
Each column of points in this plot represents correction index values from several transducers
for one token.
The horizontal axis shows time in seconds.
Here we see an increase in variability of values of the tongue body correction index
that was associated with loosening of the transducer toward the end of the run.
Tongue body data from this point on are suspect, so they are discarded.
This figure shows an example of ema data.
The symbols indicate locations of transducers on the upper and lower lips, the tongue blade,
the tongue body, and the lower incisors for multiple repetitions of the vowel o, spoken
in a carrier phrase.
The panel on the right shows data collected while a subject was holding a 1 centimeter
bite block between the molar teeth on one side during the second half of the experiment.
In the bite block condition, the lips and tongue are raised with respect to the mandible
to compensate for its lowered position.
This plot shows the same tongue body data from before the bite block and with the bite
block.
During the bite block condition, there is increased token to token variability of tongue
positions in the direction normal to the vocal tract midline.
EMA systems can provide large quantities of accurate data on the movements of midsagittal
points on vocal tract structures.
Since there are potential sources of measurement error which can't be controlled for in an
absolute sense, it's important to calibrate and use such systems very carefully in order
to assure accuracy of the data.
The challenge is then shifted from data collection to designing useful experiments and analyzing
results in the most productive and insightful way.
Thank you.
Of these, only the jaw and lips can be observed fairly easily and non-invasively because their
activity is externally visible.
This video presents several experimental applications for examining these structures using OptiTrack,
a state-of-the-art optical tracking system that accurately measures three-dimensional
position of infrared light emitting diodes attached to the structures of interest.
The primary component of the OptiTrack system is the camera bar which may be mounted vertically
or horizontally as shown here.
The bar contains three fixed focus lenses and each lens consists of a single axis element
that detects more than 2,000 positions along its axis of orientation.
Onboard algorithms increase the accuracy of the system to at least a few hundredths of
a millimeter depending on experimental conditions and the type of motion.
Three-dimensional marker positions are calculated from the known position and orientation of
the three lenses.
Tool sampling rate of the system is about 3,500 hertz.
Typically we use 10 to 20 markers whose 3D positions are measured between 100 and 250
times per second.
The camera is calibrated automatically when the system is turned on, resulting in a one-meter
viewing cube at a camera-to-subject distance of about two meters.
This large viewing area means that subjects do not have to remain perfectly still.
On the other hand, rather large markers must be used to generate a consistently detectable
light source.
First we show a simple setup in which small markers are placed mid-sagittally on the lips
and larger markers on either side of the chin.
Until recently, most optical tracking studies of the lips and jaw were set up this way and
were further restricted to measuring only one or two dimensions of motion.
Clearly, this configuration captures only a fraction of the lip behavior, particularly
during vowel rounding.
Yet, comparing articulatory events, such as the peaks and valleys in the position time
series corresponding to bilabial closures and different vowels, has provided much valuable
information about articulator timing and coordination during normal and disordered productions.
Since the OptiTrack measures 3D position of as many markers as we like, we can put them
wherever we want, as shown in this next setup, which we are using to examine facial contour
and motion during speech.
Data acquisition is controlled from the PC, which stores the marker data in a separate
file for each trial.
Marker positions can be viewed in real time or the files can be played back from within
the data collection program.
For these data, stick figures are used to connect the eight lip markers and the ten
markers on the forehead, nose, cheeks, and chin.
We're going to practice now doing the Japanese vowels.
Can you start now?
Adding a profile view gives the displayed depth and enables all three dimensions to
be seen simultaneously.
Note the substantial protrusion of the entire lip structure and cheek during production
of the rounded vowels.
In addition to marker position data, optional multi-channel AD converters can be used to
record other analog and digital signals, such as muscle EMG, electropalatography, and of
course the speech acoustics.
The next demonstration shows a cheerful subject with surface electrodes already in place for
transducing EMG activity of various orophacial muscles.
A rigid head mount is used to remove the distorting effects of head motion on the lip and jaw
data.
Instead of placing markers on the chin, whose soft tissue may move independently of the
jaw, a rigid splint is fitted to the lower teeth.
We have used muscle signals like these to model the dynamics governing lip and jaw motion
during repetitive speech sequences such as the following.
The frontal view of the lip markers and the jaw bar is shown on the left.
The very uniform time series for vertical positions are shown on the right.
Real speech is not nearly so tidy.
Pam put the bobbin in the frying pan and mommy added more pumpkin parts to the boiling potato
soup.
Pam put the bobbin in the frying pan.
The greater diversity of these data points up an important limitation of optical tracking
techniques and the need for caution in interpreting different utterance types.
Throughout the repetitive speech sequence, shown again at slower speed, the jaw and lips
are the primary articulators and their behavior is rhythmic and highly coupled, as seen in
the nearly identical lower lip and jaw traces.
However, because of the phonetic diversity of the more natural utterance, the lips and
jaw are not always primary articulators and each interacts differently with the tongue,
whose motion is inaccessible to us.
With these differences in mind, we proceed to the analysis of the muscle EMG, movement
and acoustic signals.
Aesthetic signal conditioning, such as filtering, and various manipulations of the movement
data may be done using optional Optitrack software.
Data measurement and analysis, however, must be done elsewhere.
The final experimental demonstration shows the system's ability to make complex coordinate
transformations needed for rigid body analysis of the individual components of the jaw's
3D motion.
Transformers are attached to sturdy lightweight frames that span relatively large distances
and which move in unison with the head and jaw.
The system must compute two coordinate transformations in order to specify jaw motion in its own
frame of reference.
Starting at the lower left of the figure, the first transform is from the worldview
of the Optitrack to that of the head.
This transform is necessary for basic head movement correction.
The second transform puts the jaw into its own coordinate system, which remains fixed
regardless of its motion relative to the head.
The origin is determined by the experimenter.
For the study exemplified here, the intersection of the three axes is midway between the two
jaw condyles with a horizontal axis parallel to the jaw's occlusal bite plane.
The transformed jaw motion is fully specified by rotation of the three orientation angles
and translation of the three positions.
Okay, how does that feel?
Can you move around a little bit?
Okay, why don't you try the trial?
Okay, why don't you start?
Asasa, asasa, asasa, asasa, asasa, asasa, asasa.
Although humans can move their jaws through all six degrees of freedom, for example during
chewing, we have observed only three principal components of jaw motion during nonsense
productions, all of which fall within the midsagittal plane.
Shown on the right, these are horizontal and vertical translation and sagittal rotation
of the jaw condyle.
Akaka, akaka, akaka, akaka.
Compared to the asasa productions, production of K entails quite different initial orientation
angles as well as differences in the relative amounts of horizontal and vertical translation.
In summary, pre-calibration of the camera and the data acquisition software included
with the system make it easy to use.
The system has high spatial accuracy in three dimensions and adequate sampling speed for
measuring a wide range of behaviors associated with speech.
This multiple channel analog and digital recording is possible using optional components.
The ability to track many markers and calculate coordinate transforms makes head movement
correction possible as well as complex rigid body reconstructions.
These features are particularly useful in situations where subject mobility is not easily
controlled.
The following is a list of specifications that reflects our experience with our own
optotrack systems.
Newer systems may have different levels of performance.