-->

 

Controllability Evaluation Of New Control Parameter For Myoelectric Control-Type Electrolarynx

Katsutoshi OE1, Ryoya NAKAMURA1

1Daiichi Institute of Technology (Kirishima JAPAN)

INTRODUCTION

Individuals who undergo laryngectomy as a radical treatment for laryngeal cancer lose their vocal cords and their voice; this also occurs with individuals suffering from ALS (amyotrophic lateral sclerosis) when they are fitted with a respirator.

When voice, which depend on the vocal cords and laryngeal functions, is lost, individuals are deprived of speech, their most important communication tool. This often causes acute mental distress. This has prompted research on speech product substitutes (SPSs), and some SPSs are used practically. However, they have problems with regard to voice quality, articulation and intonation. For example, the electrolarynx uses an electrovibrator as their lost sound source, and this method has good features of voice continuity, sound volume and acquisition.

On the other hand, the electrolarynx has poor voice articulation because of its uncontrollable pitch frequency. To solve this problem of the electrolarynx, many researches about the control method for electrolarynx are being conducted: Goode [1], Uemi [2]. Goldstein used the electromyographic (EMG) signal for control the on-off of an electrolarynx, but the pitch frequency could not control [3]. All of these techniques focused on controlling the pitch frequency. However, these control methods did not use the laryngeal muscle signals that control pitch in normal subjects. Our research aimed to develop a novel SPS with high controllability.

We focused on using the myoelectric signal of the sternohyoid muscle (SH) to control the electrolarynx. The SH has the function of vocal cords relaxation and is activated during the utterance of low-frequency voice. Therefore, it was suggested that the pitch frequency of electrolarynx could be controlled using the myoelectric signal of the SH.

CONTROLLABLE ELECTROLARYNX

This is the flowchart of the signal processing for our controllable elctrolarynx. The input signal is separated two signal processing flows. One of the flows is “on/off control”, indicated as (a), and the other is “pitch frequency control”, indicated as (b). In the (a) part, the signal is calculated its absolute value, and its moving average. The value of moving average is compared with specified threshold, if the average is larger than the threshold, the switch is turned on and the voice is started. If the average is smaller than the threshold, the switch is turned off and the voice is stopped. In the (b) part, the signal is filtered by band pass filter, calculated its RMS (Root Means Square) value. The pitch frequency of the electrolarynx is translated from this RMS value by a some function.
Figure 1. Flowchart of signal processing for electrolarynx control.
Our myoelectric control-type electrolarynx contains an EMG electrode, a GND electrode, an electrolarynx with control input and a control unit for signal processing and driving. The myoelectric signal is measured by an EMG electrode attached to human neck surface near the SH. To get the stable signal, this control unit has the GND electrode attached to the wrist of test subject.

The important terms for muscles that are used as the control signal sauce of an artificial larynx are as follows: 1) activation at phonation (for on-off control); 2) control of vocal cords tension (for pitch frequency control); and 3) shallow location (for detection by a surface EMG eletrode).

From above-mentioned conditions, we choose the SH. The flowchart of electrolarynx is shown in Figure 1. This system is constituted from mainly two flows, (a) on-off and (b) pitch frequency flows. The on-off control was described in our previous report [4].

To control the pitch frequency, the measured EMG signal is filtered by the band pass filter, and its RMS (Root Mean Square) value is calculated. The RMS value is translated to control signal for pitch frequency, output to the electrolarynx. For unrestricted frequency control, the conversion function for the RMS value to the pitch frequency is very important. In this report, we describe that the proposal and evaluation of new control parameter.

EVALUATION OF CONTROL FUNCTION FOR MYOELECTRIC CONTROL-TYPE ELECTROLARYNX

In order to estimate the pitch frequency control function of the electrolarynx, the relationship between the pitch frequency of human’s voice and RMS value of myoelectric signal of SH was measured and evaluated its relationship from the viewpoint of the determinant coefficient and its standard deviation.

Methods

It was reported that there were the linear and quadratic relationship between the pitch frequency of the voice and myoelectric signal of SH, and it was concluded that the quadratic relationship function was better than the linear function [5]. In this report, to decide the more detailed parameter for conversion of myoelectric signal to pitch frequency, the relationship between RMS value of measured myoelectric signal and the pitch frequency of test subject’s voice was measured. The test subjects were 5 healthy male (20 to 22 years old).

Results and discussions

The calculated determination coefficients of the approximate expressions between the RMS value of EMG signal and the pitch frequency of voice and their standard deviations are shown in Table 1. From these results, it is confirmed that the R2 values of exponents and quadratic relationship function are better than that of linear function. Furthermore, the standard deviations of exponent and quadratic are smaller than that of linear, these mean that these functions have high controllability. Therefore, it is clarified that exponent and quadratic relationship functions are suitable for control functions of the electrilarynx.

Table 1.  Calculated determination coefficients (R2) of approximate expressions and their standard deviations.

Test subjects
Approximate expressions
Linear Exponent Quadratic
R2 S.D. R2 S.D. R2 S.D.
A 0.881 0.056 0.940 0.043 0.964 0.012
B 0.788 0.063 0.955 0.025 0.925 0.039
C 0.873 0.059 0.975 0.010 0.977 0.004
D 0.853 0.047 0.932 0.025 0.944 0.016
E 0.900 0.036 0.958 0.012 0.962 0.014

EVALUATION OF CONTROLLABILITY FOR MYOELECTRIC CONTROL-TYPE ELECTRORAYNX

In this section, we evaluated the capability of EMG signal as the pitch frequency control signal for the electrolarynx by using of the experimental user interface.

Methods

This is the experimental user interface for evaluation of controllability for myoelectric control-type electrolarynx. On the left side of the figure, four indicators are located. These are called “vocalized sound indicator”, the on/off and the calculated voice tone are indicated. On the right side of the figure, the matrix with six rows and five columns included the indicators of "High", "Mid" and "Low" are located. This matrix is called “intention indicator” and visualize the subject's intention.
Figure 2. Experimental user interface.
To evaluate the controllability of the frequency control, the indication of the test subject's intention and the tone from the converted EMG signal were compared and the errors were counted. The interface set-up is presented as Figure 2.

This interface is composed of vocalized sound indicator (left side) and intention indicator (right side). The intention indicator has tone indicator functions. The test subject indicates the intention indicator of “High”, “Mid” and “Low” by mouse pointer with his intention. At same time, he is conscious of generating EMG signal. The vocalized sound indicator lights up according to the generated EMG signal. In case of the difference between the left and right indicators from captured video movie was occurred, this was counted as an error (30 fps). The test subject was a 22-year-old healthy male, one of the subjects in the previous section.

Results and discussions

This is the error rate at sound tone change from the result of controllability evaluation. This figure includes three pie charts labeled “Linear”, “Exponent” and “Quadratic”. These pie charts show the timing and percentage of errors that occurred. If error is not occurred, “no-error” is shown. In the “Linear” chart, the percentage of “no-error” is 72 %, in the “Exponent”, the percentage is 94%, and in the “Quadratic”, the percentage is 72%.
Figure 3. Error rates at sound tone change.
The error rates at sound tone change are shown in Figure 3. From this figure, it is clear that the error rate approximated with the linear relationship (shown in (a)) and the quadratic relationship (shown in (c)) are higher than that of the exponent relationship (b). The no-error rate increases from 72% to 94%, it is clarified that the exponent relationship has good characteristics for the control function of the electrolarynx.

This figure shows the number of failures at keep the constant tone. This figure includes three bar graphs labeled “Linear”, “Exponent” and “Quadratic”. The number of failures in “Liner” graph is seven times, in “Exponent” graph is once, and in “Quadratic” graph is seven times, too.
Figure 4. Number of failures at keep constant tone.
The figure 4 shows the numbers in case of the subject failed to keep the constant tone. From this result, the number of failures of exponent relationship is the least of that of other two relationship functions. Therefore, it was clarified that the most difficult condition for keeping the constant tone is the linear relationship, and the easiest condition is the exponent.

From above-mentioned results, we judge comprehensively and conclude that the exponent relationship function is suitable to control the electrolarynx.

CONCLUSIONS

In this paper, we proposed the new control parameter for myoelectric control-type electrolarynx, and evaluated the controllability of new control parameter. From the results of evaluation, the follows have been concluded.

1. The exponent relationship function between RMS value of myoelectric signal and pitch frequency of vocalized sound had good determinant coefficient.

2. Using of this function, the controllability of electrolarynx was increased than other relationship functions in the viewpoint of number of errors.

3. The exponent function was suitable for pitch frequency control for myoelectric control-type electrolarynx.

REFERENCES

[1] Richard LG, Artificial laryngeal devices in post-laryngectomy rehabilitation, Laryngoscope, 1975, 85(4):677-689.

[2] Norihiro U, Toru I, Makoto T, Jun-ichi M, Proposal of an electrolarynx having a pitch frequency control function and its evaluation, Trans. Inst. Electrom. Inf. Commun. Eng., 1995 J78-DII(3):571-578.

[3] Ehab AG, James TH, James BK, Garrett BS, Robert EH, Design and implementation of a hands-free electrolarynx device controlled by neck strap muscle electromyographic activity, IEEE Trans. On Biomed. Eng., 2004, 51(2):325-332.

[4] Katsutoshi O, Toshio F, Development of the artificial larynx with neck EMG signal control, Proc. of 2010 Int. Symp. on Micro-NanoMechatronics and Human Science(127-132). Nagoya; 2010.

ACKNOWLEDGEMENTS

A part of this research was supported by JSPS KAKENHI grant numbers 26350685 and 17K01602.