Auris Nasus Larynx 34 (2007) 327–332

www.elsevier.com/locate/anl

Electrolarynx in voice rehabilitation

Hanjun Liu a,*, Manwa L. Ng b
a
Department of Communication Sciences and Disorders, Northwestern University, 2240 Campus Drive, Evanston, IL 60208, USA
b
Division of Speech and Hearing Sciences, The University of Hong Kong, Hong Kong, China
Received 6 April 2006; accepted 10 November 2006
Available online 18 January 2007

Abstract

Objective: Patients of laryngeal cancer who have undergone the surgical removal of the entire larynx suffer the loss of phonation.
Electrolarynx (EL) speech is the most commonly adopted alaryngeal phonation. However, EL speech is notorious of the sound quality being
monotonic and robotic with the lack of pitch control and the presence of the radiated noise. This paper provides a review of modalities in EL
speech as well as introducing the technologies to control the pitch and reduce the noise of the device.
Methods: Improvements of EL speech quality have been divided into two parts: improving the sound quality of EL device by applying
different enhancement algorithms to reduce the radiated and the additive noise, and implementing pitch-control function to the EL with
advanced technology.
Results: Adaptive filtering and the subtractive-type algorithms have shown to be able to reduce the noise level associated with EL speech.
And more mature technologies are showing promise to the making of a hand-free EL system producing more accurate and synchronized pitch
and voice onset control.
Conclusion: The advent of micro-technology and human-machine integration promisingly improves EL speech quality and more efficient
algorithms enhance EL sound quality. Such improvements apparently improve the intelligibility of EL speech, and thus better quality of life of
the EL speakers.
# 2007 Elsevier Ireland Ltd. All rights reserved.

Keywords: Electrolarynx; Voice rehabilitation; Speech enhancement

1. Introduction
The removal of the entire larynx as a treatment of
laryngeal cancer usually results in the loss of the ability to
produce voice and speech. Statistical data show that there are
over 600,000 laryngectomees in the world [1], and
apparently voice restoration is essential to these people.
Standard esophageal (SE) speech and tracheoesophageal
(TE) speech are two main methods used by laryngectomees
for voice rehabilitation. But due to the low acquisition rate in
SE speech (6%) [2] and the fact that as many as one-third
of laryngectomized patients find TE speech unsuitable for
anatomical or personal considerations [3], electrolarynx
(EL) phonation is the most commonly adopted form of
phonation. An electrolarynx (EL) is a battery-powered
* Corresponding author. Tel.: +1 847 491 2428; fax: +1 847 467 2776.
E-mail address: hanjun-liu@northwestern.edu (H. Liu).
device, which incorporates the internal preset pitch that can
be adjusted to meet with individual preference for male and
female speakers. Lauder [4] and Rothman [5] found that the
use of the EL was easier, produced longer sentences without
special care, and was more effective for communication in
many situations.
Since the debut of the first EL, named Sonovox, by
Wright in 1942, EL has been undergoing many modifica-
tions. In 1945, Aurex company in Chicago started producing
an EL named Aurex Neovox M-520T, setting the design
foundation of modern EL. In 1959, the transistorized EL was
developed by the Bell Laboratories [6]. Up to date, there are
several commercially-used ELs including Nu-voice, Romet,
Amplicode, Cooper-Rand, Servox, etc. The former four do
not allow pitch adjustment during speaking, and Servox only
has two preset pitch levels (high and low) with an external
tone activation switch during conversation (see Fig. 1).
There are two different types of EL: the neck-type and the

0385-8146/$ – see front matter # 2007 Elsevier Ireland Ltd. All rights reserved.
doi:10.1016/j.anl.2006.11.010
328 H. Liu, M.L. Ng / Auris Nasus Larynx 34 (2007) 327–332
Fig. 1. The neck-type electrolarynx of Servox.
intra-oral type. The neck-type EL is the most widely used
among the laryngectomees. During phonation, the hand-held
device is held against the neck approximately at the level of
the former glottis to put the sound into the oral and
pharyngeal cavities by an electromechanical vibrator. The
vibrated electronic sound source is transmitted through the
neck tissues, where the user modulates it to create speech by
movements of articulators such as the lips, teeth, tongue, jaw
and velum [6]. The particularity of the intra-oral type EL is
the path that the sound is transmitted through. With the use
of intra-oral tube, it transmits the sound into the mouth
directly. Therefore, the energy leakage of the sound is
limited and the speech quality is better as compared to the
neck-type EL. Yet, due to the inconvenience in using the
device such as articulation and sanitary problems, intra-oral
type EL is more commonly used among laryngectomees
immediately after laryngectomy or when still undergoing
radiation therapy.
Acoustic and perceptual characteristics associated with
EL phonation have been studied extensively in the English-
speaking alaryngeal population. Because the pitch and
intensity of EL speech are fixed during phonation, only few
studies focused on the acoustic perspectives [5,7,8]. The
experiments performed to characterize spectral differences
between EL and normal phonation indicate that EL speech
emitted energy peak near 500 Hz with a maximum
amplitude near 2.5 kHz, whereas normal speech displays
a energy peak near 500 Hz. Qi and Weinberg [8] also
reported that EL speech has a substantial low-frequency
deficit with an output of 30 dB lower below 500 Hz but 5–
10 dB higher above 2 kHz than that of normal speech. Most
of them focused on describing the perceptual characteristics
of EL speech, which were usually investigated in
comparison with SE or TE speech [9–16]. It is generally
agreed that EL speech is associated with lower intelligibility,
poorer listeners’ acceptability, and more serious voicing
confusions as compared to SE or TE speech, despite both are
considerably worse than normal laryngeal speech. A review
of the literature also indicates that some studies examined
the characteristics of EL speech in tone languages including
Thai, Cantonese, and Mandarin [17–25]. Compared to non-
tone language, tone languages are characterized by wider
ranges of fundamental frequency (F 0) and faster F 0 changes.
This explains the poorer performance of EL speech. Results
indicated that EL speakers were generally not able to
produce the phonemic tones at a level of proficiency
comparable to that of normal speakers. The patterns of tonal
confusion of EL speech were different from those of NL
speech. Higher intelligibility was related to level tone for EL
speech than those related to rising or falling tones. Such poor
performance was related to EL speakers’ inability to
consistently produce pitch contours comparable to those
of normal speakers due to the limitation of the instrument
itself. Although some researches focused on acoustic and
perceptual characteristics of EL speech, the literature
presents conflicting information with regard to comparisons
of EL and other alaryngeal speakers. Discrepancies in the
findings from these studies may be due to the differences in
subject sampling, recording methods, methods of analysis or
speech samples used.
2. EL with pitch-control function
With more advanced technologies, several newer gen-
erations of EL have been developed to improve the sound
quality. As indicated in previous discussion, the monotonic
and robotic sound quality associated with EL speech is due
to the lack of pitch adjustment during phonation. Pitch is
preset at a certain level before use and remains steady during
speech production. Improvements in the EL design therefore
should focus on the real-time adjustment of the pitch.
According to the method of pitch control, these ELs can be
divided into three typical types: finger-control type [26,27],
expiration-control type [28], and electromyographic
(EMG)-control type [29]. Fig. 2 shows the schematic
diagram of the EL control module with different pitch-
control methods. While some of the finger-control type ELs
are commercially available, the expiration-control type and
electromyographic (EMG)-control type ELs are still at the
experimental stage, and pending for commercialization.
Finger-control type EL can further be categorized into
two control categories. One is to control pitch by using
finger pressure directly (e.g. TruTone EL produced by
Griffin Laboratories; UltraVoice produced by The Ultra-
Voice Plus Inc.). With the use of a pressure sensitive
transducer, pitch is controlled by increasing or decreasing
pressure placed on a single button. The greater the pressure
is provided by the finger, the higher is the pitch produced.
The other consists of a denture-based intra-oral vibrator, a
wireless miniature fingertip switch, and a controller [26,27].
A denture-based intra-oral pressure sensor and a finger
pressure sensor are used for voicing control and pitch
 H. Liu, M.L. Ng / Auris Nasus Larynx 34 (2007) 327–332
Fig. 2. Schematic diagram of the electrolarynx control module.
control, respectively. To produce natural speech, the
fingertip switch produces binary commands of voicing
and accent which are coded according to the amount of the
finger pressure, and then the controller implements the pitch
generation model based on the commands. Such finger-
control type EL is characterized by simplicity and
availability, allowing for rapid pitch changes and more
natural pitch inflection. While pressure-controlled pitch is
valuable for many speakers, those who are new to using
electromechanical speech may find it distracting and
difficult to accommodate to this pitch-control way. The
precise control of pitch is still in doubt because the
movement of the finger is hard to match the pitch changes
during speaking. Poor synchronization between speech and
finger maneuvers will often result in slow speaking rate in
order to achieve high speech intelligibility. This certainly
limits the practical application in actual verbal commu-
nication.
Another new EL with pitch-control function was
developed with the use of expiration pressure from the
stoma of the laryngectomee [28]. This EL consists of three
parts: a pressure sensor that can detect expiration pressure
produced from a stoma, an electrical circuit that can convert
expiration pressure into a fundamental frequency for voice
and an electromechanical vibrator that can be attached to
the neck. This method has been proven to be useful in
improving the naturalness of EL speech. But there are still
problems that are difficult to solve, especially for the
traditional EL users. It appears to be very challenging for a
long-time user of traditional EL to grasp this method, but
relatively easy for a laryngectomee who never used EL
before or who only uses SE or TE speech. Traditional EL
users, according to previous studies, speak by holding the
breath without the use of airflow, which makes these users
unavailable to the expiration-control EL. Moreover, the
improvement of voice quality is limited because pitch
changes in a relatively narrower range and it is difficult to
find the optimal function transforming the expiration
pressure into frequency value.
Goldstein et al. [29] designed a hands-free EL triggered
by neck muscle EMG activity. Signal processing circuitry in
a belt-mounted control unit transforms EMG activity into
control signals for initiation and termination of voicing.
329
These control signals are then fed to an EL which is held
against the neck by using an inconspicuous brace. An
envelope waveform proportional to the time-averaged power
in the EMG signal is produced by the EMG processing
circuit. The fast envelope is to turn the transducer on and off
based on a controllable threshold voltage. Simultaneously,
the slow envelope is used to control the pitch by directly
modifying the frequency of the oscillator driving the EL
transducer. The results show that this EL overcomes the
inconvenience of using hand and appears to be appropriate
for daily communication. However, it is noted that, the
EMG-control EL is still limited to the on/off control with no
further report on how it is used to control pitch.
Similarly, Painter et al. [30] designed an electromagnetic
EL which can be implanted in laryngectomees. It consists of
two parts: an activator and an implantable transducer. The
activator is composed of periodic pulse generator and
wireless frequency emitter; the transducer is composed of
subcutaneous transformer, filter, and sensor. Both are
connected with wire and implanted in the neck tissue of
the laryngectomee. During phonation, the pulse generator
produces a periodic signal like other vibrating source of EL,
and passes it to the vocal tract with the use of emitter,
transformer, filtering, and sensor. Finally speech is produced
by movement of vocal organs. As compared to the neck-type
or intra-oral type EL, the most advantage of the implantable
EL is that the pitch and the intensity can be adjusted during
phonation without the use of hand. However, several factors
should be considered when using this EL among the
patients: compatibility, repellency, and duration of the
implanted materials, the size of the EL, the technique and
the cost of the surgery, etc. Meanwhile, this surgery is very
dangerous to the older patients. All these factors tend to limit
the development of the implantable EL, which is still in the
stage of lab testing.
3. EL speech enhancement
During EL phonation, some of the sounds produced by
the vibrating diaphragm are radiated directly from the
device. Poor interface with the neck and the surrounding
neck tissues may result in radiated noise which interferes
330 H. Liu, M.L. Ng / Auris Nasus Larynx 34 (2007) 327–332
with the intelligibility of the speech. In the extreme cases,
the stiff neck tissue resulting from radiation therapy may
reflect all the acoustic energy from the EL back into the
environment without propagating into the oral cavity for
articulation. This apparently fails the transmission of
meaningful words to listeners. When the mouth was closed,
the intensity of radiated noise was about 20–25 dB [6], and
this value varied over 4–15 dB across the subjects for the
same device [7]. Several factors have been found to be
related to the amount of radiated noise, including the manner
in which the device is coupled to the neck, and the vibratory
behavior, and the radiation characteristics of the neck tissue.
Data from previous reports indicated that the radiated
noise resulted in a poor intelligibility. This is especially
worse with low signal-to-noise ratio (SNR) [13]. An
important acoustic cue that distinguishes voiced and
voiceless stops is the presence of a periodic low-frequency
signal during the closed portions of voiced stops. However,
due to continuous operating of the EL during phonation, the
periodic radiated source noise would exist even during the
closure portion of both voiced and voiceless stops, resulting
in confusion of voicing of stops. In addition, EL has peak
noise energy between 0.4 and 1 kHz and between 2 and
4 kHz, which may contribute to the unnaturalness and poor
quality of EL speech [7].
In order to limit the influence of the radiated noise on EL
speech, Norton and Bernstein [31] attempted to improve EL
speech by changing the driving signal of the vibration
source. Defined as the ratio of the spectrum of the pressure
signal measured at the lips and the spectrum of the input
signal of a shaker vibrating against the neck, the neck
frequency response function (NFRF) of the normal subjects
was used to modify the output of EL so that the spectrum of
the sound exciting the vocal tract resembled that of a natural
human glottal source. By applying a 1 in. thick foam shield
around the EL, they found that listeners judged the modified
EL speech as sounding as more natural. However, the
soundproof shield failed to provide effective noise isolation
as well as increasing the size of EL and counterbalancing the
damping provided by the hand holding the device.
Due to the limitation of acoustic shielding techniques,
researchers have considered the use of signal processing
techniques for EL speech enhancement. Fig. 3 shows the
schematic diagram of the typical EL speech enhancement
system. The literature reveals that two methods have been
put forth to improve the perception of EL speech [32–37].
The first method is adaptive filtering [32–34]. This method
reduces noise by removing the noise components of the
primary input signal with the help of the reference input
signal based on second-order statistics. However, many
other noise components in the primary input signal that
depend on the noise reference signal have to be estimated
based on higher-order statistics. Not only does this decrease
the efficiency of the algorithm but it also affects the noise
reduction ability. The second method of EL speech
enhancement is the subtractive-type algorithm which
Fig. 3. Schematic diagram of the electrolarynx enhancement system.
includes power spectral subtraction (PSS) [35], magnitude
spectral subtraction (MSS) [36], and spectral subtraction
with quantile based noise estimation (SS-QBNE) [37].
Based on the assumption that speech and additive noises are
uncorrelated and that speech signal is a slow-varying
process, subtractive-type algorithms estimates the back-
ground noise and then to subtract the estimation value from
the noisy speech in the frequency domain [38]. These
subtractive-type algorithms are efficient in reducing the
radiated noise because it is a periodic signal determined by
the device and accurate estimate of it is very simple.
However, the subtraction parameters of these algorithms
used for EL speech enhancement are fixed and cannot be
adapted to the different levels of noise.
More importantly, previous researches focused on the
radiated noise reduction of the EL under the situation of
absolutely quiet in the lab. However, EL speech commu-
nication often takes place in the noisy environment. The
low-energy EL speech is easier to be masked by the different
environment noises, in turn decreasing the intelligibility of
EL speech. Therefore, a few efforts have been put into the
enhancement of EL speech in the case of the additive noise.
Considering the particular characteristics of EL, this
investigation mainly focused on improving EL speech in
electronically mediated environments without the need of
talking face to face, including when using the telephone,
addressing public gatherings, or in any situation in which
electronic media could reasonably be employed. Adaptive
filtering algorithm based on independent component
analysis is put forward to reduce the babble noise of the
EL speech [34]. Another two methods of improved spectral
subtraction incorporated with the perceptual weighting filter
[39] and auditory masking model [40] were also applied to
the reduction of additive noise in EL speech, in which a high
flexibility of subtraction parameters variation is offered.
 H. Liu, M.L. Ng / Auris Nasus Larynx 34 (2007) 327–332
These two subtractive-type algorithms consider the auditory
masking properties of the human ear, which allows finding a
better tradeoff between the amount of noise reduction, the
speech distortion and the level of residual noise in a
perceptual sense. Furthermore, in order to overcome the
limitation of the noise estimation in the subtractive-type
algorithms, we are developing a new method based on the
adapted wavelet packet transform to improve EL speech,
and the results of the pilot study is inspiring.
4. Summary
As mentioned above, considerable researches have been
conducted to investigate the practical and theoretical
improvements of EL speech. With the development of the
state-of-the-art technology, we have reasons to believe that
significant advances in improving the speech quality of EL
speech in two aspects. The first aspect is the ability of the EL
to adjust pitch and intensity real-time during phonation. The
EMG-control EL will likely be adopted for several reasons.
First of all, the method of pitch control using EMG signals is
the most natural, and it is accomplished without the help of
intentional finger or expiration pressure manipulation.
Secondly, the response time of EMG is faster than the
finger or expiration pressure during phonation, which is
expected to solve the synchronization issue between pitch
control and speech production. Finally, surgery similar to the
implantable EL is not required and such EL can be used
easily without the dependence of the hand. The other aspect
of technological advances in EL development is the speech
enhancement system of EL used for electronically mediated
environments. Considering the rapid development of
information technology, communication via electronic
media becomes more prevalent. It is essential to improve
the quality of EL speech to make the laryngectomees live
with comfort and convenience, thus better quality of life.
Now we are developing this system in which different
algorithms of speech enhancement are embedded. They can
be selected by using a switch, based on different noise
conditions, to obtain a high intelligibility for better
understanding. With the use of Digital Signal Processing
technology, we are trying to realize the enhancement
function with a microprocessor and embed it into a
telephone, microphone, or other electronic media. With
the development of efficient enhancement methods, the
quality of EL speech will be extensively improved for better
understanding and thus higher quality of life.
References
[1] Hirokazu S, Takahashi H. Voice generation system using an intra-
mouth vibrator for the laryngectomee. MS thesis. Japan: The Uni-
versity of Tokyo; 2000.
[2] Hillman R, Walsh M, Wolf G, Fisher S, Hong W. Functional outcomes
following treatment for advanced laryngeal cancer. Part 1. Voice
331
preservation in advanced laryngeal cancer. Part II. Laryngectomy
rehabilitation: the state-of-the-art in the VA system. Ann Otol Rhinol
Laryngol 1998;107(Suppl 172, Pt 2):1–27.
[3] Karen C, Joel M. Utilization of microprocessors in voice quality
improvement: the electrolarynx. Curr Opin Otolaryngol Head Neck
2000;8:138–42.
[4] Lauder E. The laryngectomee and the artificial larynx—a second look.
J Speech Hear Disord 1970;35:62–5.
[5] Rothman H. Acoustic analysis of artificial electronic larynx speech.
In: Seikey A, editor. Electroacoustics analysis and enhancement of
alaryngeal speech. Springfield, IL: Charles Thomas; 1982. p. 95–118.
[6] Barney HL, Haworth FE, Dunn HK. An experimental transistorized
artificial larynx. Bell Syst Tech J 1959;38:1337–56.
[7] Weiss MS, Yeni-Komshian GH, Heinz JM. Acoustic and perceptual
characteristics of speech produced with an electronic artificial larynx.
J Acoust Soc Am 1979;65:1298–308.
[8] Qi Y, Weinberg B. Low-frequency energy deficit in electrolaryngeal
speech. J Speech Hear Res 1991;34:1250–6.
[9] Knox AA, Anneberg M. The effects of training in comprehension of
electrolaryngeal speech. J Commun Disord 1973;6:110–20.
[10] Hyman M. An experimental study of artificial-larynx and esophageal
speech. J Speech Hear Disord 1955;20:291–9.
[11] McCroskey RL, Mulligan M. The relative intelligibility of esophageal
speech and artificial larynx speech. J Speech Hear Disord 1963;28:
37–41.
[12] Shipp T. Frequency, duration, and perceptual measures in relation to
judgments of alaryngeal speech acceptability. J Speech Hear Res
1967;10:417–27.
[13] Holly SC, Lernman C, Randolph K. A comparison of the intelligibility
of esophageal, electrolarynx, and normal speech in quiet and in noise. J
Commun Disord 1983;16:143–55.
[14] Williams SE, Watson JB. Differences in speaking proficiency in three
laryngectomy groups. Arch Otol 1985;111:216–9.
[15] Gandour J, Weinberg B. Perception of contrastive stress in alaryngeal
speech. J Phonet 1982;10:347–59.
[16] Gandour J, Weinberg B. Production of intonation and contrastive stress
in electrolaryngeal speech. J Speech Hear Res 1984;27:605–12.
[17] Gandour J, Weinberg B, Petty SH, Dardarananda R. Vowel length in
Thai alaryngeal speech. Folia Phoniatric Logo 1987;39:117–21.
[18] Gandour J, Weinberg B, Petty SH. Voice onset time in Thai alaryngeal
speech. J Speech Hear Disord 1987;52:288–94.
[19] Gandour J, Weinberg B, Petty SH, Dardarananda R. Tone in Thai
alaryngeal speech. J Speech Hear Res 1988;53:23–9.
[20] Ching TY, Williams R, Van Hasselt CA. Communication of lexical
tones in Cantonese alaryngeal speech. J Speech Hear Res
1994;37:557–71.
[21] Ng M, Kwok C, Chow S. Speech performance of adult Cantonese-
speaking laryngectomees using different types of alaryngeal phona-
tion. J Voice 1997;11:338–44.
[22] Ng M, Lerman J, Gilbert H. Perceptions of tonal changes in normal
laryngeal, esophageal, and artificial laryngeal male Cantonese speak-
ers. Folia Phoniatric Logo 1998;50:64–70.
[23] Ng M, Gilbert H, Lerman J. Fundamental frequency, intensity, and
vowel duration characteristics related to perception of Cantonese
alaryngeal speech. Folia Phoniatric Logo 2001;53:36–47.
[24] Liu HJ, Wan MX, Wang SP, Niu HJ. Aerodynamic characteristics of
laryngectomees breathing quietly and speaking with the electrolarynx.
J Voice 2004;18(4):567–77.
[25] Liu HJ, Wan MX, Wang SP. Features of listeners affecting the
perceptions of Mandarin electrolaryngeal speech. Folia Phoniatric
Logo 2005;57:9–19.
[26] Takahashi H, Nakao M, Kikuchi Y, Kaga K. Alaryngeal speech aid
using an intra-oral electrolarynx and a miniature fingertip switch.
Auris Nasus Larynx 2005;32(2):157–62.
[27] Takahashi H, Nakao M, Okuas T, Hatamura Y, Kikuchi Y, Kaga K. A
voice-generation system using an intra-mouth vibrator. J Artificial
Organs 2001;4:288–94.
332 H. Liu, M.L. Ng / Auris Nasus Larynx 34 (2007) 327–332
[28] Uemi N, Ifukube T, Takahashi M, Matsushima J. Design of a new
electrolarynx having a pitch control function. In: Proceedings of the
IEEE international workshop on robot and human communication;
1994. p. 198–203.
[29] Goldstein EA, Heaton JT, Kobler JB, Stanley GB, Hillman RE. Design
and implementation of a hands-free electrolarynx device controlled by
neck strap muscle electromyographic activity. IEEE Trans Biomed
Eng 2004;51:325–32.
[30] Painter C, Kaiser T, Fredrickson JM, Karzon R. Human speech
development for an implantable artificial larynx. Ann Otol Rhinol
Laryngol 1987;96(5):573–7.
[31] Norton RL, Bernstein RS. Improved laboratory prototype electrolar-
ynx (LAPEL): using inverse filtering of frequency response function of
the human throat. Ann Biomed Eng 1993;21:163–74.
[32] Espy-Wilson CY, Chari VR, MacAuslan J, Walsh M. Enhancement of
electrolaryngeal speech by adaptive filtering. J Speech Lang Hear Res
1998;41:1253–64.
[33] Espy-Wilson CY, Chari VR, Huang CB. Enhancement of alaryngeal
speech by adaptive filtering. ICSLP Proc 1996;2:764–7.
[34] Niu HJ, Wan MX, Wang SP, Liu HJ. Enhancement of electrolarynx
speech using adaptive noise cancelling based on independent
component analysis. Med Biol Eng Comput 2003;41(6):
670–8.
[35] Cole D, Sridharan S, Moody M, Geva S. Application of noise
reduction techniques for alaryngeal speech enhancement. In: Proceed-
ings of the IEEE TENCON-97, vol 2; 1997. p. 491–4.
[36] Pandey PC, Bhandarkar SM, Bachher GK, Lehana PK. Enhancement
of alaryngeal speech using spectral subtraction. In: Proceedings of the
DSP 2002, vol 2; 2002. p. 591–4.
[37] Pratapwar SS, Pandey PC, Lehana PK. Reduction of background noise
in alaryngeal speech using spectral subtraction with quantile based
noise estimation. In: Proceedings of the seventh world multiconfer-
ence on systemics, cybernetics and informatics; 2003. p. 408–13.
[38] Boll SF. Suppression of acoustic noise in speech using spectral
subtraction. IEEE Trans Acoust Speech Signal Process 1979;27:
113–20.
[39] Liu HJ, Zhao Q, Wan MX, Wang SP. Enhancement of electrolaryngeal
speech based on auditory masking. IEEE Trans Biomed Eng 2006;53:
865–74.
[40] Liu HJ, Zhao Q, Wan MX, Wang SP. Application of spectral sub-
traction method on enhancement of electrolarynx speech. J Acoust Soc
Am 2006;120:398–406.