Summary
Using both an electronically controlled resistance simulating the variations in neck resistance caused by vocal fold vibratory patterns and live measurements of vocal fold contact area, it is shown that the Glottal Enterprises EG series electroglottograph (EGG) has an inherent background noise that is less than that of the Laryngograph/Kay Elemetrics EGG units by roughly 15 to 20 dB, and less than the noise in a F-J Electronics EGG unit by at least 11 dB. The measurements presented support the claim that the lower noise in the Glottal Enterprises units make them usable with almost all men, women, and children over the age of four.

Introduction
An electroglottograph is a device that transduces the small variations of transverse electrical resistance of the neck at the level of the larynx that are caused by the variations in vocal fold contact as the folds vibrate during voice production. To make this measurement, a small AC voltage of about one volt rms or less, and usually at two to three megahertz, is applied to the surface of the neck via a pair of gold-plated electrodes.

In the early 1970s, interest in the use of the electroglottographs a non-invasive means for monitoring vocal fold vibratory patterns increased when the Laryngograph company introduced a form of the EGG that produced usable waveforms with most adult subjects and older children. Their EGG (also known variously as an 'laryngograph' or 'electrolaryngograph') featured circular electrodes having concentric rings that were believed to focus the field of sensitivity of the unit within the neck, to reduce the level of noise artifacts. Proper electrode position was determined by moving the electrodes on the neck during a prolonged vowel or voiced sonorant consonant articulation until a maximum EGG signal level was obtained. This, of course, assumed a relatively stationary larynx position during the testing procedure, though some laryngeal movement could be tolerated once the nominally optimal electrode position was located, and also assumed a larynx height low enough with respect to the mandible that the electrodes can be moved above the larynx during the positioning maneuver (which may be problematic with some women and children). Since the Laryngograph EGG has been marketed for over a decade in the U.S. by Kay Elemetrics, sometimes under Kay's brand name, we refer to it here as the Laryngograph/Kay electroglottograph.

However, the Laryngograph/Kay EGG, though a significant technological advance for its time, still did not give usable waveforms for many subjects, most noticeably women (and some men) with a considerable amount of fatty and/or muscular tissue covering the larynx, and younger children. Another EGG introduced more recently by F-J Electronics appeared to present similar problems, namely, excessive noise for some subjects and difficulty in monitoring the correct placement of the electrodes on the neck.

During the 1990s, Glottal Enterprises developed a new EGG that employed a dual-channel configuration that allows the user to continuously monitor the location of the larynx with respect to the electrodes and provides to the user an unambiguous front panel indication of the proper electrode location. [M. Rothenberg, A Multichannel Electroglottograph, Journal of Voice, Vol. 6., No. 1, pp. 36-43] For subjects with a high larynx position, this feature also meant that positioning of the electrodes did not require the user to move the electrodes above the level of the larynx. When made available on the front or rear panel, the value of the display could also be recorded along with the EGG signal.

In addition to its unique ability to indicate the position of the glottis with respect to the electrodes, the new Glottal Enterprises EGG was designed with a high signal-to-noise ratio by using modern state-of-the-art electronics and carefully matching the electronic properties of the unit to the electrical impedance of the neck. The result was an EGG that provided a usable waveform for almost all subjects, and for most subjects provided a waveform that had little or no visible random noise component.

To make possible objective measurements of signal-to-noise ratio, Glottal Enterprises has also developed the LS-1 Larynx Simulator (a description is available on the Glottal Enterprises website at www.glottal.com/calibrator.html ). The LS-1 can be connected to the EGG unit in place of the electrode leads and produces a well-defined periodic variation in electrical resistance of a magnitude of 0.1% peak-to-peak about a mean value of resistance of 150 Ohms. This mean value was chosen to match an average or typical transverse neck resistance at the operating frequency range for electroglottographs.

The peak-to-peak variation of 0.1% was chosen to represent an average value for the voices of adult subjects. Voices we have measured in our research and development efforts have consistently yielded values between roughly 0.03% and 0.3% during normal voice production in non-breathy speech.

Measurements of Signal-to-Noise Ratio

Method
As illustrated in Figure 1, a LS-1 Larynx Simulator was inserted in place of the neck electrodes to provide a simulated neck resistance of 150 Ohms, with a calibrated periodic change in resistance of 0.1%. These values were chosen to match typical values for the human neck with normal voice production. (In our measurement procedure for the Glottal Enterprises EGG, the LS-1 was connected to the EGG's electrode cable input jack through a length of cable similar to the cable used for the electrodes. For the other two EGGs tested, the electrode cable was left in place and the LS-1 was connected to the electrodes.) The signal-to-noise ratio was obtained by measuring the peak-to-peak output voltage on the screen of an oscilloscope, with and without the periodic change in the emulated neck resistance. The ratio of these two voltages is the signal-to-noise ratio (S/N), and can be expressed either as a multiple (as X10) or in dB (as 20 dB for a multiple of X10). All testing was performed in a typical office/laboratory environment with some electronic instruments and computers nearby, but no unusual radio frequency noise sources present. The setup in Fig. 1 was also used for the F-J electroglottograph.

Since the EGG output voltage with no periodic change in simulated neck resistance (emulating the electrodes in place, but no voice production) was a random noise signal, the peak-to-peak value could only be determined approximately. It was estimated by visually estimating a voltage range that incorporated approximately 90% of the positive and negative voltage peaks. In addition, low frequency components of the noise, below about 50 Hz, were ignored, since in practice they would be masked by the physiologically based low frequency noise present in all EGG signals, and also because they could theoretically be removed by appropriate (linear phase) digital filtering. These various measurement factors could cause an uncertainty in the resulting measured S/N of one or two dB.

The Laryngograph/Kay unit required a special connection to the LS-1, since it had electrodes with two concentric elements, an inner disc and a ring. On one electrode, the ring is not connected, and serves no apparent purpose, and thus was neglected in connecting the LS-1. However, the ring on the other electrode is active and must be considered when connecting the LS-1. The manner in which the LS-1 is connected is shown in Figure 2. We have assumed that the complex electrical field patterns caused by the concentric electrodes can be represented by two elements of neck resistance, Rt and Rs, which must be both emulated to properly employ the LS-1.

The neck resistance element Rt is the transverse electrical resistance (including the current flow path through the larynx), and it is emulated by connecting the LS-1 between the center discs of each electrode.

The neck resistance element Rs is a surface resistance between the center disc and the ring of the same electrode. To estimate Rs, we placed a 3 MHz voltage between the disc and ring and measured the resulting current when the electrode was placed on the skin of the neck. (Three megahertz is the operating frequency of the Laryngograph/Kay unit.) A thin coating of electrode jelly was used to maintain a good contact with the skin. This procedure resulted in a value of between 50 and 80 Ohms, depending on the pressure applied to the electrode.

We also checked to see if variations in this resistance would significantly affect the measured S/N, by obtaining S/N values for a range of values of Rs. Varying this element (Rs) was found to vary S/N such that a peak occurred at roughly 150 to 200 Ohms. Thus to measure the maximum performance that might be obtained with a Laryngograph/Kay unit, the surface resistance Rs was emulated by connecting a fixed resistance of 180 Ohms between the disc and ring of the appropriate electrode. (Parenthetically, this variation of sensitivity with Rs might explain why the instructions for the Laryngograph/Kay unit do not specify the use of electrode jelly; a poor contact at the active electrode ring might actually improve performance, or at least compensate for any deleterious effects of poor contact at the center disc.)

Results
Glottal Enterprises EGGs: Six randomly chosen Glottal Enterprises EG2-PC units from current production, carrier frequency 2 MHz, were tested as in Figure 1. All yielded a S/N factor of X180, plus or minus 20%, or roughly 45 dB, within plus or minus about two dB. The measured S/N for these EGG units was therefore between about 43 and 47 dB for a 0.1 peak-to-peak signal level, assuming an average transverse neck resistance of about 150 Ohms, the value emulated by the LS-1. (This test justifies the manufacturer's rating for the EG2-PC, or the electrically similar EG-2, as yielding an output having a random noise component at least a factor of 100, or 40 dB, below the signal produced by a variation of transverse electrical neck resistance of 0.1%.)

Laryngograph/Kay EGG: One Laryngograph/Kay EGG, carrier frequency 3 MHz, approximately 10 years old and carrying the Laryngograph brand name, was tested as in Figure 2. The value of Rs was chosen to be 180 Ohms. Since the bandpass range of the Laryngograph/Kay unit was measured by us to be greater than that of the Glottal Enterprises EG2 unit (roughly 9 kHz vs. 6 kHz for the EG2), a single-pole low pass filter was used at the output of the Laryngograph/Kay unit to match the two units and allow a direct comparison of signal-to-noise ratio. This test yielded a S/N factor of approximately X10 for a signal level representing a of 0.1% variation in transverse neck resistance. This is equivalent to a S/N of approximately 20 dB.

If we allow an error of up to 2 dB in the S/N measurements for the Kay unit, our tests indicate that the S/N ratio for the Glottal Enterprises EGGs (45 dB plus or minus 2 dB) was greater than that for the Laryngograph/Kay EGG by at least 21 dB (43 dB - 22 dB). This is equivalent to a noise level for the Laryngograph/Kay unit that is greater by at least a factor of X11.

F-J Electronics EGG: One F-J Electronics EGG, model EG80, approximately 3 years old, was tested as in Figure 1. The carrier frequency is 400 kHz for this unit. A problem in obtaining comparable noise measurements for this unit was the band-pass of the F-J electroglottograph, which at 3500 Hz was almost half that of the Glottal Enterprises units and almost one-third that of the Laryngograph unit. This lower value for the upper frequency limit has the effect of lowering the noise level by roughly 2.5 dB to 3.5 dB compared to the other units tested (while also limiting the waveform fidelity during periods of rapid change). To allow meaningful noise level comparisons between units, the noise reading we found for the F-J unit was increased by 3 dB, to approximate the increase in noise that would be found if the F-J unit had a bandwidth increased to a value similar to those of the other two units tested, say 7 kHz. (For Gaussian white noise, doubling the noise bandwidth increases the noise amplitude by 3 dB.)

Because of the presence of some carrier signal component at the output for low signal levels, it was necessary to add a low noise low-pass filter before the signal was sent to the oscilloscope. An Ithaco model 4302 filter in the low-pass mode was used for this purpose. The -3 dB cutoff point for this filter was set at 10 kHz, which was sufficiently above the 3.5 kHz cutoff frequency of the unit so as not to appreciably affect the random noise level or signal accuracy.

We found that for our standard 0.1 % variation in simulated neck resistance, as applied by the LS-1 Artificial Larynx, the signal from the F-J unit was approximately 45 times the noise level, which is equivalent to a S/N ratio of about 33 dB. If we reduce this by 3 dB to account for the narrower bandwidth of the F-J unit, the S/N is about 30 dB. Allowing an error of plus or minus 2 dB, we get a S/N of 28 dB to 32 dB. This is 11 to 19 dB less than the S/N ratio of the Glottal Enterprises units (43 to 47 dB). Thus, according to our measurements, the effective noise level of the F-J unit was at least a factor of 3.6 times greater (11 dB) than the noise level of the Glottal Enterprises unit and slightly less than the Laryngograph/Kay unit tested.

Note: The manual for the F-J unit specifies a S/N ratio of 47 dB at a signal level representing a 4 % variation of electrical neck resistance, with an average electrical neck resistance of 100 Ohms (slightly less than the 150 Ohms used in the LS-1 Larynx simulator). This is equivalent to a S/N of 15 dB for a 0.1 % variation. This specification is approximately 18 dB less than our measurements indicate (33 dB with no compensation for bandwidth), and would indicate a noise level almost 10 times the level we found. Though the problem could be in the calibration of the LS-1, it is also conceivable that there was a typographical error in the F-J manual, and the reference signal level was only 0.4 % instead of 4 %.

Performance comparisons with human subjects: The noise level differences reported above were generally substantiated in a number of informal tests with five human subjects (3 men, 2 women and a child). Though some small differences were occasionally noted, the noise level of the Glottal Enterprises EGG was consistently the lowest for all subjects tested.

A typical result comparing the Glottal Enterprises and Laryngograph EGG units is shown in Figure 3. Both waveforms were recorded from the same 4 year, 3 month old boy, producing two similar vocalizations during the same recording session. The small-size electrodes available from Glottal Enterprises were used for its EGG (29 mm OD, with a 27.5 mm diameter active area, as compared to the 32 mm diameter of the Laryngograph/Kay unit electrodes and 24 mm diameter of the F-J electrodes). Using the LS-1 as a calibration reference, the peak-to-peak variation in transverse electrical neck resistance for this subject and this type of vocalization was estimated from the upper (Glottal Enterprises EGG) waveform to be as 0.062 %. (The peak-to-peak amplitude of the upper waveform was 0.62 times that of the EGG output with the LS-1 connected in place of the electrodes.) The waveforms shown in the figure were photographed from the screen of a storage oscilloscope. From the time scale (2 ms/div), the fundamental frequency (F0) of the upper waveform was 307 Hz, while that of the lower waveform was 278 Hz. For both recordings, considerable care was taken to position the electrodes so as to yield the best possible waveform while the electrodes were held in place manually by the experimenter.

For a transverse neck resistance variation of 0.062%, our model of the Laryngograph/Kay electrodes predicts a S/N of 6.2, or 16 dB. However, the signal in the lower trace of Figure 3 is more closely a factor of 10 times the peak-to-peak noise level, or 20 dB. Thus this result generally supports the S/N estimates made using the LS-1 Larynx Simulator, but indicates that the simulation may have yielded noise levels about 4 dB too high for this subject.

It can also be noted parenthetically from Figure 3 that, when using the Glottal Enterprises EGG, it was possible to obtain a clear EGG trace from a child as young as 4 years old. The limiting factor was the tolerance of the child for the electrodes and his patience during the measurement procedure.

Conclusions
It is difficult to get a precise comparison between the signal-to-noise ratios of the EGGs tested, because of the unusual lead configuration of the Laryngograph/Kay unit, the difficulty in assuming an optimal electrode position for the Laryngograph/Kay and the F-J units, and the difficulty in getting untrained subjects to repeat a vocalization at the same F0, voice quality and loudness. However the results reported here support the conclusion that, for most subjects, the Glottal Enterprises EG2 units provided a noise level at least about 16 dB less than the Laryngograph/Kay unit tested, and 12 dB less than the F-J unit.

Because of their low noise level, the Glottal Enterprises electroglottographs should provide a usable waveform (at least 20 dB above the random noise level) for voices yielding a peak-to-peak variation in transverse electrical neck resistance as low as about 0.01 %. For an equivalent waveform fidelity, the Laryngograph/Kay unit appears to require a variation in transverse electrical neck resistance of at least 0.06 %, and possibly higher for some subjects, even given some low pass filtering to reduce the signal bandwidth to a value similar to that of the standard EG2. The F-J unit would require a variation in transverse electrical neck resistance of at least 0.04 % for the same noise level if signal bandwidths were equalized.

In view of the many potential applications of an electroglottograph that can be used conveniently with almost all subjects, it would seem desirable to extend these results to other subjects, preferably using voice professionals in order to obtain more reliable results, and to other commercial EGGs. Some values of electrical neck resistance other than 150 Ohms should also be tested in the simulation, and preferably more than one unit tested for each model. The test procedures used by each manufacturer for their ratings should also be requested and reported. It would also be preferable for such tests to be carried out by a researcher not having a vested interest in a particular brand, as does the author.
---------------------------------------------------
Technical Note: The Laryngograph and Glottal Enterprises units are powered by internal rechargeable batteries, with the Glottal Enterprises unit having a dual-battery system that allows one battery to be charged while the other battery is in operation, to provide continuous operation. The F-J unit requires both an internal replaceable battery and power from the game/joystick jack of a computer's audioboard.

*************
January 4, 2002
Martin Rothenberg, Ph.D.
President, Glottal Enterprises
Professor Emeritus of Electrical and Computer Engineering, Syracuse University

Figure 3. EGG waveforms in VFCA polarity (movement up indicates an increase in vocal fold contact area) from a 4 year, 3 month old boy producing a sustained vowel /a/. The horizontal time scale is 2 ms/div.

Top: Using a Glottal Enterprises EG2-PC electroglottograph. [F0 = 307 Hz]
Bottom: Using a Laryngograph/Kay electroglottograph. [F0 = 278 Hz]