INTELLIGIBILITY OF ELECTROLARYNX SPEECH

USING A NOVEL HANDS-FREE ACTUATOR

Brian Madden, Mark Nolan, Edward Burke, James Condron and Eugene Coyle

Department of Electrical Engineering Systems, Dublin Institute of Technology, Dublin, Ireland

Keywords: Laryngectomy, Electro-larynx, Intelligibility, Pager motor, Hands-free.

Abstract: During voiced speech, the larynx provides quasi-periodic acoustic excitation of the vocal tract. In most

electrolarynxes, mechanical vibrations are produced by a linear electromechanical actuator, the armature of

which percusses against a metal or plastic plate at a frequency within the range of glottal excitation. In this

paper, the intelligibility of speech produced using a novel hands-free actuator is compared to speech

produced using a conventional electrolarynx. Two able-bodied speakers (one male, one female) performed a

closed response test containing 28 monosyllabic words, once using a conventional electrolarynx and a

second time using the novel design. The resulting audio recordings were randomized and replayed to ten

listeners who recorded each word that they heard. The results show that the speech produced using the

hands-free actuator was substantially more intelligible to the majority of listeners than that produced using

the conventional electrolarynx. The new actuator has properties (size, weight, shape, cost) which lends itself

as a suitable candidate for possible hands-free operation. This is one of the research ideals for the group and

this test methodology presented as a means of testing intelligibility. This paper outlines the procedure for

the possible testing of intelligibility of electrolarynx designs.

1 INTRODUCTION

A total laryngectomy is typically performed due to

cancerous growths in the neck. Not only does it

involve complete removal of the larynx, but also the

trachea is disconnected from the pharynx and

redirected through a permanent aperture in the front

of the patient's neck (the tracheostomy or stoma), as

shown in Figure 1 (National Cancer Institute, 2010).

During voiced speech air is expelled from the lungs

which provides the power source for excitation of

the vocal tract, either through laryngeal phonation

(voiced sounds), turbulence in a vocal tract

constriction (unvoiced sounds) or a mixture of both.

In each case, the actual speech sound produced

varies according to the configuration of the vocal

and nasal tracts. Post total laryngectomy, normal

speech is impossible because the conventional

sources of vocal tract excitation are absent. The total

laryngectomy procedure deprives the patient of their

primary channel of communication. Since the loss of

speech has an enormous impact on quality of life,

speech rehabilitation is an important aspect of

recovery following this surgery.

Figure 1: Redirection of airways following a TL.

1.1 Electrolarynx

Laryngectomees, as a part of their rehabilitation, are

trained to communicate with as much ease as

possible. Most are trained to use oesophageal or

tracheo-oesophageal speech. For a minority, these

channels of communication are not possible. For this

remaining group, the most common form of

communication is to use an external speech

prosthesis. This is a mechanical larynx which uses

an electromechanical actuator, i.e. the electro-larynx.

The modern electro-larynx was invented by Harold

Barney in the late 1950s (Barney and Madison,

1963). It is a hand-held, battery-powered device

which incorporates a transducer that generates

265

Madden B., Nolan M., Burke E., Condron J. and Coyle E..

INTELLIGIBILITY OF ELECTROLARYNX SPEECH USING A NOVEL HANDS-FREE ACTUATOR.

DOI: 10.5220/0003167902650269

In Proceedings of the International Conference on Bio-inspired Systems and Signal Processing (BIOSIGNALS-2011), pages 265-269

ISBN: 978-989-8425-35-5

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

mechanical pulses at a single frequency within the

natural range of the human voice. The transducer

uses an a coil-magnet arrangement that vibrates

against a diaphragm when the output of an electrical

oscillator is applied to its winding. The device is

pressed against the mandible and this vibrates the

pharynx which in turn resonates the air in the vocal

and/or nasal tract. The vibrations are formed into

speech by the articulators of the upper vocal tract.

Research to date has focused on the

improvement of the quality of speech produced by

the electro-larynx. Some significant contributions

have been made by (Houston et al. 1999) who

developed an electro-larynx which used digital

signal processing to create a superior quality of

sound. (Shoureshi et al. 2003) used neural-based

signal processing and smart materials to improve the

sound created. (Liu et al. 2006) and (Cole et al.

1997) focused on removing the buzzing sound

created by the transducer. (Uemi et al. 1994)

developed a system that utilized measurements from

air pressure that was obtained from a resistive

component that was placed over the stoma to

maintain the electrolarynx’s fundamental frequency.

(Ma et al. 1999) used cepstral analysis of speech to

replace the electrolarynx excitation signal with a

normal speech excitation signal.

Despite all the acoustic improvements these

studies have shown, they have however, been

performed in isolation and have been deemed to be

difficult to implement into the existing technology.

Therefore, the basic concept of design which was

first introduced by Barney et al. in the 1950’s

remains the same to this day. It has been shown that

up to 50%-66% of all laryngectomees use some

form of electrolarynx speech (Gray et al. 1976) and

(Hillman et al. 1998): either as a method of

communication for speech rehabilitation post-

surgery or as a reliable back-up in situations where

esophageal or trcheo-esophageal speech is proving

difficult.

1.2 Speech Intelligibility

When determining the intelligibility of a speech

signal, it is important to choose a suitable linguistic

level at which to make measurements. Is it necessary

to measure the accuracy at with which each phonetic

element is communicated in order to assess whether

each word is identifiable. It is also necessary to

investigate whether the communication of a sentence

is clear.

This type of linguistic dismissal can introduce an

additional difficulty in that individual human

listeners will ultimately differ in their capability to

make use of these linguistic constraints. Even

though it may be deemed necessary to assess the

utility of a particular channel in order to convey the

meanings of real spoken utterances, listeners will

inevitably vary in their capacity to comprehend the

speech, depending on their own linguistic ability.

Many speech intelligibility tests consist of either

phonetic unit, which are composed into: nonsense

syllables, words which are used in isolation or in

short sentences spoken in one breath for comfort

(Crystal et al 1982) and (Mitchell et al. 1996).

An issue that arises through the use of nonsense

syllables is that many listeners could require training

in order to be able to identify the component

phonetic units, and they may be confused by

phonemes which don't compare well with the

spelling e.g. there, their, they’re. Therefore by

limiting listener reply’s to real words thus allowing

them to respond in ordinary spelling. This can

however introduce other difficulties: firstly, that

varying listeners may possess differing degrees of

familiarity with the words that are being used;

secondly, that some words are memorable and

having heard a word once, some listeners may be

biased in their usage of a particular word another

time.

A possible solution to these problems includes

the formulation of multiple word lists of reasonable

difficulty, allowing a listener to be used within a test

more than once. Another option is to create tests

consisting of closed response sets, making every

listener needing to make the matching choices about

the word which is under test.

(Egan et al. 1948) pioneered one of the first lists

of words for an intelligibility test in 1948. He

created the list by using the concept of “phonetic

balance” which meant that the relative frequency of

the phonemes in the word lists corresponded to the

relative frequency of phonemes in conversational

speech. He constructed 20 lists containing 50

monosyllabic words and his intention was to balance

average difficulty and range of difficulty throughout

the lists whilst ensuring that the phonetic units that

were present were represented equally.

2 METHODS

2.1 Novel Actuator Design

This design consists of a simple pager motor (which

is typically found in a mobile phone) attached to a

thin piece of high-density polyethylene (HDPE) by

BIOSIGNALS 2011 - International Conference on Bio-inspired Systems and Signal Processing

266

an aluminium support. When a current is sent

through the motor, it causes the off-centred cam to

rotate causing an unbalanced centrifugal force. The

motor in this design has a minute amount of play

within the support and the HDPE section, thus

causing a vibration that is resonated through the

plastic. The thinner the plastic material is, the better

the resonance becomes and resulting in a more

efficient transfer of vibrations into the user’s neck.

Figures 2 and 3 show the novel design for this

concept and the region of a user’s neck where it is

envisioned that the final device will be attached.

Figure 2: Novel motor design.

Figure 3: Region where device is attached.

2.2 Testing Parameters

Two able-bodied speakers (i.e. non-laryngectomees),

1 male and 1 female, were chosen as participants

and they received some basic pre-training in the use

of an electrolarynx. They were instructed prior to

recording to locate the point on their neck which

produced the best resonance and thus the best

sounding output (also known as the “sweet spot”).

They were asked to hold their breath and maintain it

held during each audio recording. Once while using

the commercially available Servox electrolarynx and

once while using the novel pager motor design.

A randomized sample of 30 words from one of

Egan’s list was taken and used for the intelligibility

test bed. As there were only two participants, the 30

words were reduced to 28 so as to have a number of

recorded samples of each device per speaker that

was easily devisable by 2 i.e. 7 random words for

each device and for each speaker.

The participants were instructed to sit upright in

a chair and in order to keep the subject’s posture

constant during testing; their foreheads were

supported in a head rest, figure 4. The height of the

subject’s seat was adjusted until an angle of 100

degrees from the chin to the torso was achieved. The

microphone was then positioned 15cm away from

the subject’s mouth, figure 4.

Figure 4: Experimental setup using head rest.

2.3 Test Methodology

A combined microphone and preamplifier (Maplin

KJ44X) was used to record the vocalization audio

signals. The microphone was connected to a

National Instruments 6023E 12-bit analog-to-digital

converter (ADC). The ADC was set to a sampling

frequency of 44.1 kHz on all channels. Before

testing, the pre-amplified microphone was calibrated

using a Brüel & Kjær 2231 Sound Level Meter, at a

distance of 15cm from a constant audio signal

source. The audio intensity was adjusted and the

output voltage from the microphone preamplifier

was compared with the corresponding recorded

Sound Pressure Level (SPL). A virtual instrument

(VI) was created in LabVIEW which streamed in the

data from the microphone through the ADC to the

computer. The incoming data stream was broken

into recordings of 5 second segments. The VI gave a

visual display of the recordings and they were saved

as an .lvm file in a folder on the computer after

recording of each utterance. Figure 5 illustrates a

block diagram of the test set up.

INTELLIGIBILITY OF ELECTROLARYNX SPEECH USING A NOVEL HANDS-FREE ACTUATOR

267

Figure 5: Block diagram of test set up.

2.4 Test Methodology

After the test was completed, each labview

measurement file was converted into a waveform

audio file format file for convenience so as to be

able to create randomized audio playlists containing

the 28 recorded utterances and played to a listener

on an audio player. Each waveform audio file format

file was normalized to an audio level of -19dB on

Cooledit Pro Version 5, with the out of band peaks

selected as having no limits (i.e. not clipped).

Prior to creating the playlists, a formulated

organisation of each one was arranged by taking a

randomised selection of the 28 words for each

individual playlist. Each word was then matched as

shown in tables 1 and 2. For example, if the first

word on the playlist was “leave”, the recording for

“leave” was extracted from the recordings for each

speaker and assigned to the word on the list.

Table 1: Playlist order for listener 1.

Number Speaker Device

1 to 7 1 EL

8 to 14 1 Pager

15 to 21 2 El

22 to 28 2 Pager

Table 2: Playlist order for listener 2.

Number Speaker Device

1 to 7 1 Pager

8 to 14 1 El

15 to 21 2 Pager

22 to 28 2 El

Every second playlist was arranged so as to

alternate the device being heard first, table 2. This

was done so as not to create a listener bias towards

the Servox electrolarynx as it was deemed that it

could possibly take a number of recordings until the

listener began to understand what to concentrate on.

3 RESULTS

The results for 90 percent of the listeners indicated a

greater intelligibility of the utterances which were

spoken using the hands-free pager motor design. The

results were tabulated subjectively using a 0 to 1

scoring system. This quantative analysis is presented

in table 3.

Table 3: Quantative analysis of point scoring system used.

Recorded

utterance

Interpretation of

utterance Result Score

Leave Leave

√

Which Witch X 0.75

Towel Dowel X 0.5

Rude Ruth X 0.25

Exam Sap X 0

An average of the overall intelligibility score of

all ten listeners for both speakers resulted in the

novel hands-free pager design being 30% more

intelligible that the commercially available Servox -

59% intelligibility scored for the pager and 30%

intelligibility for the Servox. Even when the

utterances from the pager motor were in the first

group heard by the listeners number 2, 4, 6, 8 and

10, they were still greater than the next group of

seven spoken using the Servox electrolarynx; the

pager motor design averaging at 78% intelligible and

the Servox averaging at 55% intelligible. Figure 6

illustrates the mean score obtained for both devices

by each listener, where listeners are represented on

the x-axis and their cumulative scores are

represented on the y-axis.

Figure 6: Mean scores of both devices for each listener.

(Blue represents EL/Red represents Pager)

4 CONCLUSIONS

This aim of this was to compare the intelligibility of

speech that was produced using a novel hands-free

BIOSIGNALS 2011 - International Conference on Bio-inspired Systems and Signal Processing

268

actuator compared to that of speech produced using

a conventional electrolarynx. The results illustrated

that the speech produced using the novel actuator

were substantially more intelligible to all the

listeners than that produced using the conventional

electrolarynx. The limitations of this study are in the

fact that only two speakers and ten listeners were

used, a much larger variety of subjects (including a

large number of electrolarynx users) would be

required to create the random database of

monosyllabic words for listeners to subjectively

quantify. Further to this future intelligibility studies

will not be limited only monosyllabic words and will

be expanded to intelligibility of standardized

phrases. Future paradigms of the future scoring

system are to be explored as many listeners found it

difficult to decipher certain words which have

meanings with and without a vowel at the beginning

e.g. “Bout” and “About” or “Wake” or “Awake”.

Initial results are encouraging and further work in

the provision of a second iteration of a hands-free

facility which will be tested using this methodology.

REFERENCES

National Cancer Institute (2010). What you need to know

about cancer of the larynx. Retrieved July 20, 2010,

http:// www.cancer.gov/ cancertopics/ wyntk/

larynx/page2

Barney H. L, Madison N. J., 1963. Unitary Artificial

Larynx. United States Patent Office. Patent No;

798,980.

Houston K. M., Hillman R. E., Kobler J. B., Meltzner G.

S., 1999. Development of sound source components

for new electrolarynx speech prosthesis. IEEE Proc.

International Conference on Acoustics, Speech, and

Signal Processing Vol. 4, Issue, 15-19 Page(s):2347 -

2350 vol.4.

Shoureshi R. A., Chaghajerdi A., Aasted C., Meyers A.,

2003. Neural-based prosthesis for enhanced voice

intelligibility in laryngectomees. IEEE/EMBS Proc. 1

Int. conference on neural engineering, Capri Island,

Italy.

Liu H., Zhao Q., Wan M., Wang S., 2006 Enhancement of

electrolarynx speech based on auditory masking. IEEE

transactions on Biomed. Eng. Vol. 53, No. 5.

Cole D., Sridharan S., Mody M., Geva S., 1997

Application of noise reduction techniques for

alaryngeal speech enhancement. IEEE TENCON –

Speech and image technologies for computing and

telecommunications. 491-494.

Uemi, N., Ifukube, T., Takahashi, M., & Matsushima, J.

(1994). Design of a new electrolarynx having a pitch

control function, IEEE Workshop on Robot and

Human, 198–202.

Ma, K., Demirel, P., Espy-Wilson, C., & MacAuslan, J.

(1999). Improvement of electrolarynx speech by

introducing normal excitation information.

Proceedings of the European Conference on Speech

Communication and Technology (EUROSPEECH),

Budapest 1999, 323–326.

Gray, S., & Konrad, H. R. (1976). Laryngectomy:

Postsurgical rehabilitation of communication.

Archives of Physical Medicine and Rehabilitation, 57,

140–142.

Hillman, R. E., Walsh, M. J., Wolf, G. T., Fisher, S. G., &

Hong, W. K. (1998). Functional outcomes following

treatment for advanced laryngeal cancer. Part I–Voice

preservation in advanced laryngeal cancer. Part II–

Laryngectomy rehabilitation: The state of the art in the

VA System. Research Speech-Language Pathologists.

Department of Veterans Affairs Laryngeal Cancer

Study Group. Annals of Otology, Rhinology and

Laryngology Supplement, 172, 1–27.

Crystal, T. H., & House, A. S. (1982). Segmental

durations in connected speech signals: Preliminary

results. Journal of the Acoustical Society of America,

72, 705–717.

Mitchell, H. L., Hoit, J. D., & Watson, P. J. (1996).

Cognitive–linguistic demands and speech breathing.

Journal of Speech and Hearing Research, 39, 93–104.

Egan, J. (1948). Articulation testing methods.

Laryngoscope, Vol. 58(9), pp. 955-991

INTELLIGIBILITY OF ELECTROLARYNX SPEECH USING A NOVEL HANDS-FREE ACTUATOR

269