The Speech Hearing and Language Research Centre (SHLRC) at Macquarie University has a unique Physiology Laboratory, integrating several physiology devices. Most physiology devices are targeted for one aspect only, and as such are difficult to integrate and synchronise with other devices. As an example, a college in the UK has one computer which measures articulator movement, and another computer that measures palate contact and audio, and a third computer that controls them. The data is them loaded into another computer for analysis. Synchronisation, and the time to transfer the data, is equivalent to the data session, which means each session is double the time required, and messy to analyse.
The system at SHLRC has all the systems, integrated, synchronised, and
data can be analysed immediately after gathering, all on one computer.
The data that can be simultaneously recorded consist of:
Audio, 2 channels
Articulography, 10 sensors
Electropalatography
Airflow, 3 channels
Electromyography, 4 channels
Anything else that we haven't thought of yet.
AUDIO
The system has 2 channels for audio and speech input. Each channel has a 16 bit A/D converter capable of gathering data up to 20 kilohertz bandwidth using a sample rate of 40 khz. Normally speech is gathered only on the Left channel at 20khz sample rate. Both audio input channels have a low noise floor of -80dbm and a maximum level of +16dbm, using balanced transformer coupled inputs. This arrangement is for noise free recording of AC signals only (no DC offset problems) with a flat frequency response from 20hz to 20khz.
The system has 2 channels for audio and speech output. Each channel has a 16 bit D/A converter capable of outputting data at up to 20 kilohertz bandwidth using a sample rate of 40 khz. The system is capable of a maximum 100khz sample rate. Normally speech is output only on the Left channel at 20khz sample rate. Both audio output channels have a low noise floor of -80dbm and a maximum level of +16dbm, using balanced transformer coupled outputs. This arrangement is for low noise reproduction of AC signals only (zero DC offset) with a flat frequency response from 20hz to 20khz.
A tone generated by the synchronisation circuit is attached to a small speaker, so that it is picked up by the microphone, and serves as an audio marker.
There are also 2 channels of Intensity, which are used for smoothing
of the audio channels and producing a profile of the speech signal. These
are seldom used now, as most computer software shows the speech envelope.
These date from the days of the Mingograph which was a fast (sprayed ink
type) pen recorder.
AUDIO CONTROL
The audio input and output is controlled from a panel in a small rack on the desk. It allows 2 channel audio input from several sources, including microphone, tape recorders, or any other balanced input. There is a facility to measure and adjust RMS level and Peak level to ensure adequate level without overload. The audio output channels also have similar output capabilities. There is a monitor facility to enable listening to the audio either by speaker or headphone, both channels or either channel.
ARTICULOGRAPHY
There are two articulography systems, and either can be used, or both used simultaneously, but it would be impractical, and usually the superior device is used.
MOVETRACK
This is a device for measuring physical movement of objects and made by Botronic of Sweden. Small sensors can be glued to the face, lips and tongue. When the sensors move in an electromagnetic field, they pick up signals from 2 transmitters, the signal strength being proportional to the distance from the transmitter. The 2 transmitters are placed on a small helmet, worn by the subject. The 2 transmitters represent the X and Y directions of the mid sagittal plane through the subjects oral cavity. Since the transmitters are each energised by a different frequency, and the sensor coils pick up both, the distance from each transmitter can be plotted in real time as a distance from the transmitter, and thus represents tongue and lip movement. There are 4 channel amplifiers which each produce an X and Y signal for that channel sensor.
It is critical to adjust the X and Y transmitters to 90 degrees and that they remain fixed for the duration of the measurement. This is a weakness of this device. If a sensor tilts during speech (tongue rolling, glue lifting), less signal is picked up, which is represented as a move away from the transmitter, and constitutes an error.
ARTICULOGRAPH
This is another device for measuring physical movement of objects, similar
to the Movetrack, made by Carstens of Germany. It uses the same principles
as the Movetrack, but uses 3 transmitters arranged at 120 degrees and permanently
fixed to a rigid helmet. The sensor
electrodes are attached in a similar manner, but pick up 3 signals.
These signals are analysed and with some mathematics, produce an X Y coordinate,
but with automatic correction for tilt! The results form this device are
more reliable than the Movetrack due to the rigid helmet and automatic
tilt correction. The original I.E.E.E. 488 computer interface was removed
and a new one designed in SHLRC. This interface also included a signal
monitor facility and a synchronisation marker. There is a calibration cradle
and program, used to make sure all new sensors are correctly set up. Each
sensor has individual calibration data as each one can be different due
to manufacturing tolerances. A special program was written to calibrate
the sensors. There is a paper about this interface, presented at the Speech
Science and Technology Conference 2000. Link.
ELECTROPALATOGRAPH (E.P.G.)
To measure tongue contact on the roof of the mouth, a custom made false
palate is inserted into the oral cavity. The palate has 62 small studs
arranged in a fixed pattern. These studs receive a signal from a hand held
electrode, when touched from by the tongue. The output from the palate
is processed by the Reading University EPG 3+ device and is output
as a series of digital signals, and variable sample rates. The EPG is controlled
from the interface. There is also a synchronisation circuit designed and
built by SHLRC that marks the data, so that the time waveforms can be aligned
and checked for data integrity.
AIRFLOW
There are 3 channels of pressure amplifiers which are setup to measure air parameters. These are connected to 3 pressure transducers. The 3 differential transducers are heavy and are attached to the chair headrest, so that their weight doesn't cause subject fatigue. Each transducer is connected by small plastic air pipes to the actual sensor. Two of the sensors are for airflow and consist of a piece of gauze that is arranged so that the air passes through it. The gauze causes a slight resistance, and consequently a pressure differential, which is conveyed to the transducer. The sensor is placed in the air path of the subject, for measuring the Nasal airflow or Oral airflow. A third sensor is placed inside the oral cavity to measure actual pressure, as compared to ambient pressure.
The are 2 transducer amplifiers made by Teledyne, and a third one which was built by SHLRC. They are essentially the same and can be adjusted for gain and offset. There is also 3 lowpass filters with switchable frequencies, which are used to smooth the signal and remove any noise. A switchable highpass filter and 4 additional amplifiers are available if the signal requires further enhancement.
ELECTROMYOGRAM (E.M.G.)
To measure muscle activity, electrodes can be attached to the facial muscles. These are a small non-invasive surface coil that sticks to the skin with adhesive tape, instead of the older "harpoon" or "fish hook" (barbed style) electrodes, that pierce the skin to enter the muscle directly. The surface type have to placed carefully, to ensure good electrical pickup from the muscle, without interference from other muscle groups. They are very sensitive and may require repositioning for reliable results.
There are 4 channels of E.M.G.. input using equipment from Neotrace. Each channel has an amplifier which has switchable gain and switchable low and high cut filtering. Each channel has an integrator to enable smoothing of the signal, with a switchable time constant.
DIGITAL MONITOR
There is a digital meter which can be switched to several inputs to allow precise setting of DC offsets to zero. This is useful fro setting up airflow amplifiers.
CARDIAC MONITOR
There is a large screen monitor which allows individual physiology signals to be monitored. These appear as slow moving horizontal lines on a screen. Each of the 8 channels has a position control, so that signals can be logically grouped and compared. Each channel also has an individual gain control. This is useful for examining any physiology signal during setup, or checking for errors and system failures.
PATCH PANEL
This allows access to the 16 inputs to the slow speed A/D converter. These can be individually patched to any channel, although the reserved channels are generally used. The audio outputs can be monitored here as well. The Cardiac monitor channels can be patched to any of the channels to easily change the monitoring function.
SHLRC INTERFACE
All of the physiology signals are analog signals, and are attached to
several A/D converters which have various numbers of channels and sample
rates. The audio has a SHLRC designed computer card, that has a single
high quality 16 bit converter for each audio channel, and can digitise
audio, or reproduce it. There are 2 of these. The other physiology signals
are applied to a 12 bit A/D converter that has 16 channels, and this is
used for the airflow, EMG, and Movetrack data. There are another 2 identical
12 bit A/D converters that are used to digitise 30 channels of data from
the Articulograph. All these converters are controlled from a SHLRC designed
and built microprocessor device. This device functions as a buffer and
performs a First In First Out function (FIFO) to gather all the physiology
data, buffer it, and them send it to the host computer, where it is stored
and analysed. The FIFO is programmable, so that any combination of physiology
channels, audio, and buffering can occur. The sample rates of the audio
and physiology can be programmed to any rate (in 1 hz increments), not
just a fixed set. However, rates of 20khz for the audio and 200hz for the
physiology are commonly used. There is a further digital interface on the
FIFO that controls the EPG and collects the data from it simultaneously.
There is a program on the host computer written by SHLRC which calibrates,
collects, formats and saves the data. The data can the be examined by programs
like XWAVES or EMU. There is a paper about this interface, presented at
the Speech Science and Technology Conference 1986. Link.
DEFINITIONS
SAMPLING THEORY: Nyquist says that you must sample at twice the highest
frequency you are measuring for accurate data representation, otherwise
frequency errors occur, often called aliasing.
TRANSDUCER: This a device to convert physical energy to electrical
energy, which means that a physical movement can be converted to an electrical
signal. For example, a microphone is a transducer that converts variations
in air pressure (sound) to electricity (a sine wave like signal). Another
common transducer is a speaker which does the exact opposite to a microphone.
FILTER: A device for removing some unwanted frequency component from
a signal, often available as a "high cut" or "low pass" filter.
INTEGRATOR: A device to "add up" signals, often used for smoothing
signals.
A/D: Analog to Digital converter, a device to digitise analog
signals
D/A: Digital to Analog converter, a device to digital numbers into
a real world signal.
PHYSIOLOGY WEB SITE
http://www.shlrc.mq.edu.au/~calaghan
ACKNOWLEDGEMENTS
Thank you to all those that have helped in this project, especially Chris Callaghan and Sallyanne Palethorpe from S.H.L.R.C.