Marek  Roland - Mieszkowski
Digital Recordings - Advanced R & D
5959 Spring Garden Rd., Suite  1103
Halifax, Nova Scotia B3H-1Y5,Canada

Andrzej Czyzewski  &  Bozena  Kostek
Technical University of Gdansk
Sound  Engineering  Department
ul.Narutowicza 11/12,80-952 Gdansk,Poland

Copyright 1990-1994, Digital Recordings
Halifax, Nova Scotia, Canada


New models and a new approach to the stuttering
disorder were developed.  Elements of this new
theory were applied to explain research results and
many features of the stuttering disorder. They were
the basis for the design and construction of a new
electronic device as a method for elimination or
reduction of stuttering. DSA is based on an advanced
DSdP (Digital Sound Processing) of speech signal in
the auditory feedback loop (Fig.1). The device uses
the most modern digital signal processing circuitry,
designed according to the algorithms developed by
the authors, to elicit the desired response
characteristics of the feedback signal.

Authors believe that stuttering is a physiological
disorder, in most cases of a neurological nature.
Stutterers become nervous, because they stutter, not
as believed by some, they stutter since they are
nervous. DSA is aimed to correct stuttering in the
case of "classical stutterers", who are characterized
by usual patterns of stuttering, have no problem with
speech synthesis and do not stutter during
whispering. In some cases subjects outside this group
can also benefit from this technology, but results are
less predictable. With DSA many stuttering people
can speak fluently (or more fluently) in any fashion
and at any rate.

      |                                    |
      |    ****  Photograph of DSA ****    |
      |                                    |

Fig.1. Digital Speech Aid - III-d generation model

Long-term effects were tested on several patients.
Due to the relaxing and reassuring effect of DSA, the
base-line stuttering level decreases.  The
effectiveness of the device over time appears to
remain the same for removal of the "natural level" of
stuttering. In a sense, DSA is working as a prosthetic
device, similar to a hearing aid.

Origins of Stuttering and Standard Treatment

In many cases stuttering was believed to be caused
by psychological disorders and nervousness of the
person. This stigma still exists in the large part of the
society and large part of the medical and health
community. Many stutterers themselves believe in this
very strongly. Outcome of this believe is not only
great suffering on the part of the person, due to the
fact of being labeled as "weird", "nervous" and
"psychologically unstable" but also very often wrong
approach to the treatment of the disorder.

Treatment offered by the Speech Pathologists
involves various techniques to slow down the speech,
coordinate speech production with breathing, change
the way of speaking and pronouncing words etc. It
also involves some counselling and relaxation
therapy, which very often overlaps with work and
input from the Psychologist. These techniques work
to certain degree and results depend very much on
the particular case. Also they work often better in the
clinical setting than in the real world, where person
cannot concentrate as much on speech production.
And unfortunately many of these techniques require
a conscious effort on the part of the stutterer. Many
people give up the speech therapy because in some
cases they feel that fluent, but unnatural sounding
speech is worse then stuttering itself. It is estimated
that about 5% to 10% of stutterers are receiving some
form of the therapy ( indicative of the current
treatments effectiveness) [1,2].

New models and electronic speech aids

If stuttering is a physiological disorder (of neurological
nature in most cases), telling a person to control it
does not make much sense. It is like telling a person
with faulty vision to take off glasses and to concentrate
to see better [2].

This approach to the problem was supported by many
researchers. Among others  Fairbanks [3] noticed the
role of the feedback monitoring in speech production.
According to this  theory  the outgoing speech
movements are controlled by a cybernetic system
which depends  upon feedback based on sensory
information for the maintenance of the right
performance. Besides the sensory, or kinesthetic-
tactile mechanism, there is another feedback loop
mechanism acting at the acoustic level. This auditory
feedback plays a very important role in the speech

The Wiener's cybernetic theory presented in 1948
caused the appearance of many theoretical models
dealing with stuttering origin and mechanism. These
theories are based on the assumption that the speech
output is returned to the central integrating system
through the airborne side tone, bone conducted side
tone, tissue connected side tone and kinesthetic-
tactile sensors on both sides of the body. Stromsta [4]
noticed in 1962, that the auditory feedback signals  in
these different channels arrive in different moments
and that the resulting information received by the
brain becomes very complex. Subsequently, many
authors [5,6] dealt with many possible sources of
distortions in the feedback systems used to monitor
speech. They include asynchrony of feedback signals
that arrive in the right and the left hemispheres and
also differential delays in bone and airborne feedback

Consequently, it is evident that the following question
need to be raised: how could we modify the distorted
feedback loop in order to improve the speaker's
performance ? As  it is known since the fifties, the
strong effect occurs when auditory delays are
introduced when a person is speaking[6,7,8]. While
many explanations proposed by Stromsta [4] may
have some partial validity, it is worth to notice that it
is also possible that the reason for the observed fluency
is that the patient is trying to suppress monitoring of
his own delayed voice, because the delayed speech
is disturbing. Another known effects are masking
noise (MAF)  [8]  and frequency-altered auditory
feedback (FAF)  [5].

However, nobody noticed the role of discorrelating of
the signals as the more universal approach to the
modification of the auditory feedback loop. The
mentioned discorrelation concerns the signals arriving
to the sensory system of the brain from the tissue
(afferent channels) and acoustic loops. Looking at the
known electronic techniques of stuttering suppression
one can notice, that there exist an obvious link
between the degree of the expected discorrelation
and the efficiency of the method. For example: total
masking (through the use of the very loud noise or
caused by the deafness) proves to be very effective.
This result is not surprising, because it causes a total
discorrelation through the decay of one of the signals.
The frequency alteration performed in the auditory
feedback loop (FAF) also seriously discorrelates the
sensed signals. Delayed feedback (DAF) causes less
discorrelation, so it  is not surprising that the
performance of this method is worse in many cases
[9,10]. Moreover, some methods are disadvantageous
because of their unpleasant effects. For example the
echo is slowing speech and the masking noise is
annoying and even may be dangerous to the hearing

Consequently, the best results with known methods
were found by the authors when using FAF or FAF
combined with small delays in DAF [9,10]. Interesting
question arises : are there other methods leading to
the discorrelation of the signals received by the
speech monitoring system  that are more efficient and
are less distorting the speech received by the
stutterer ?. The expertise gained by the authors in the
domain of signal processing allow them to positively
answer above question. Consequently, various new
techniques were investigated during the thorough
experiments conducted at the Sound Engineering
Dept. of the TU Gdansk, among others by the master
degree candidates [9,10].

Some of the newly proposed DSdP and filtering
techniques are realizable only in the digital domain,
so there was a need to design and to construct the
proper digital processing platform to perform these
experiments - namely DSA.  The mentioned
experiments are in progress now, and authors' new
results will be published soon.

Performed experiments allow to draw a conclusion
that the digital filtering and similar techniques using
the discorrelation techniques effect very strongly the
speech production process. Depending on the
parameters of the algorithm it is possible to cause
both the increase and the decrease of the stuttering
frequency. Parameters and the results depend very
strongly on the stutterer case. Consequently, one of
the DSA's software version was provided with 256
settings of filter parameters selectable by the user or
more precisely, by the investigators implementing the
device at this stage. The prepared program options
allow one to match the filtering parameters to the
individual needs of the patients. However, such
experiments are very time consuming process  and
optimization of proposed algorithms is  not an easy
task. Hypothesis that the discorrelation effect may be
considered as a common feature of the most effective
methods of electronic stuttering elimination was
confirmed during these initial  tests.

Based on these experiments, the new concepts of the
DSdP algorithms for DSA are being implemented.
They also open a new  interest  in  search for new,
more  effective methods of alteration of the auditory
feedback loop to reduce  stuttering.

Digital Speech Aid (DSA)

DSA - III generation model (Fig.1) is a small
11cmx6cmx3cm (LxWxH), sophisticated, electronic
device with 256 different program settings and 5
independently variable parameters. All known and
many new algorithms can be easily implemented in
the existing hardware, due to the fact that DSdP
(Digital Sound Processing) software is stored in
EPROM, which could be easily changed and
reprogrammed. DSA uses a microphone and pair of
earphones. It operates on batteries. The device is
relaxing and non-disturbing. With DSA a person can
speak  in any fashion and at any rate. DSA is most
effective in the case of "Classical Stutterers" who
consist about 80% -> 90% of the stuttering population.
Significant improvement or total fluency is observed
in about 40% -> 60%  of "Classical Stutterers". Rest
of them also improve to various degrees. Device was
and is still tested in real life, not artificial
laboratory situation. Improvements were observed in all
situations : in the office, at home, on the telephone,
during public meetings, presentations, good and bad
days,  etc.  Improvement is instant, however, we
observe increase in effectiveness during the first 2 ->
6 weeks (this is consistent with other similar
observations and could be explained in author's
opinion on the bases of neural networks, which could
be used as the model of inner-working of the auditory
system). After that it seems to remain the same. In
majority of cases there is a carryover effect - person
remains more fluent for 2 hours -> 2 weeks after using
DSA. Significant improvement in self-esteem and
self-confidence are observed. People like to use DSA
and say that it is relaxing. Many people also indicate,
that they feel, that they cannot stutter (in authors'
opinion, speech production/monitoring system
indicates to them that conditions for fluent speech are
fulfilled). Long term effects seem to support authors'
theory and expectations - DSA  is still effective (same
level) after being used for 10 months [2].

Facts about stuttering and implications for treatment

About 4% of children and 1% of adults stutter.
Stuttering changes with age. People stutter to a
varying degree and in different ways. Often people
stutter on particular sounds. Often rate of the
stuttering varies for given individual (depending on
various factors). Stuttering usually depends on
language used by the person. Males stutter 3 times
more often than females. Stuttering starts in the early
age and in some cases goes away at later age. In
many cases food and alcohol (or other chemicals)
change stuttering - better or worse. In many cases
exercise and physical activity change stuttering. In
many cases stress changes the rate of stuttering -
better or worse.

Becoming suddenly deaf leads to total fluency. With
shadow speech (whispering)  or choral speech (with
other person) - majority of cases is fluent (90% ?).
When singing or talking in noise (cafeteria, bar,
music) - majority of cases is fluent (90% ?).  Lowering
or increasing pitch of the ones voice, assuming
foreign accent and slowing down the rate of speech
production also results in increased fluency .

Amplification or attenuation of the voice , delay of the
voice in the range 1->100 ms, white or other types of
noise, frequency shifting of the voice in the range -1 -
> +1 octave, reverberation of the voice, combination
of the above increases fluency.

From above facts it is obvious that hearing plays very
important role in speech production and control. It is
also clear that stuttering is caused by physiological
disorder, neurological in nature in most of the cases.
Speed of propagation of neural signals seems to play
important role (facts and experiments), lower
frequencies are more important than higher (facts and
experiments), vocal tone is very important (facts and
experiments). Stuttering seems to be correctable by
the processing of sound (facts and experiments).

How to correct stuttering via Signal Processing ?

Methods of stuttering correction could be divided into
three broad categories :
*  Masking- for example various types of noise, etc.
   make signal unusable for control in the Speech
   Control System (SCS). SCS relies on other afferent
   channels in this case (Fig.2).
*  Non-Masking - discorrelation of the signals - for
   example DAF, FAF, Reverberation. This seems to
   be better, since signal is not as disturbing and it is
   comprehendible as voice by the higher levels of
   Speech Synthesis System (SSS), therefore helping
   in this synthesis (Fig.2). However SCS is probably
   not using this signal for the control (servo) purposes.
*  Correction - signal is shaped via DSdP (Digital
   Sound Processing) in such a way as to correct for
   deficiencies and at the same time make it still
   acceptable by SCS for the control (servo) purposes.
   This is the preferred way of correcting, since it will
   be more effective and pleasant to use by the
   stutterer.  Authors' hypothesis, based on our theory
   and experiments is that it is possible to use this type
   of systems in certain cases of stuttering. It should be
   also possible in certain cases to gradually adjust
   parameters of DSdP algorithms to do retraining of
   the neural network associated with the speech
   production / auditory system. It is expected that in
   some cases of stuttering the damage or pathology
   is on the level of neural network programming and
   in others on the level of neural network structure (or
   "hardware"). Currently authors are engaged in
   applications of neural networks for DSdP.  Further
   tests and experiments are in progress.

Hypothesis about SCS :
In stutterers the auditory signal is used by SCS, but
from time to time the voice signal is not being
accepted leading to prolongations and other observed
stuttering effects (Fig.2). By manipulating signal via
DSdP,  one can obtain auditory feedback which will
be on one hand acceptable by SCS  for control
(servo) purposes and on the other hand will lead to
correction of speech production. This should in turn
lead to fluency. Hopefully this could be done for all
sounds produced by the stutterer. Also one would
hope, that this correction will be working over the
whole range of variability (stress, alcohol, etc).

                  |  SSS  |    SSS - Speech Synthesis System
                  ---------    SCS - Speech Control System
                     A |       SPS - Speech Production System
                     | V         C - Cochlea
        |---------|  SCS  |  <---------------------------------|
        |         ---------  <----------------|                |
        |                                     |                |
        V                                     |                |
    ---------  Air conducted Sound  -----------------------    |
    |       | --------------------> |   |                 |    |
    |  SPS  | Bone conducted Sound  | C | Auditory System |    |
    |       | --------------------> |   |                 |    |
    ---------                       -----------------------    |
        |                                                      |

        |                           -----------------------------
        |-------------------------> |  Other Afferent Channels  |

Fig.2. Block diagram of the speech production system (ASCII version).

Introduction of some kind of processing - ear plugs,
amplification, equalization, filtering, DAF, FAF, DSA,
noise, etc. changes auditory feedback. SCS is
comparing the remembered signal (by the already
formed and "fixed" neural network) with the produced
signal and is correcting its shape. If SCS cannot do
this, it will get stuck (stuttering effect) - system is
trying to overcome a problem and is trying to do this
over and over again. Since most of the SCS work is
under a subconscious control, the stutterer has very
little control over it.



Performed experiments and research allow the
authors to draw a conclusion that the digital filtering
methods using the discorrelation techniques effect
very strongly the speech production process.
Depending on the parameters of the algorithm it is
possible to cause both the increase and the decrease
of the stuttering frequency. However, the optimal
algorithm parameters depend very strongly on the
particular stutterer case.

Our short-term and long-term experiments with DSA
and various DSdP algorithms indicate, that auditory
conditions play paramount role in the speech
production. Fact, that some people can speak at any
rate and in any fashion with certain forms of DSdP,
shows, that stuttering disorder should be correctable
similarly to hearing impairment (with hearing aids) at
least in certain cases.

In authors opinion, any realistic experiments should
involve wearable and portable device, such as DSA in
order to allow testing in various real-life environments
and situations. Also long-term performance can be
investigated only with the portable device such as

Authors believe, that we are very close to find new
more efficient methods to eliminate the stuttering
disorder through the electronic speech aids. The
progress in this domain may be brought by the
implementation of the modern DSdP technology.

Our hope is that scientists from different fields will
join forces together in order to advance our knowledge of
this disorder and its treatment. Without this approach
this progress will be as slow as in the last several


[1]   Hermes Electronics, internal report " The Digital
      Speech Aid", Halifax, Nova Scotia, Canada,
      December 1993.

[2]   Roland-Mieszkowski M., book "Career
      Assassination", Halifax, Nova Scotia, Canada, to
      be published.

[3]   Fairbanks , G. (1955),. Selective vocal effects of
      delayed auditory feedback, Journal of Speech and
      Hearing Disorders, 20.

[4]   Stromsta C. (1990). Delays associated with certain
      sidetone pathways, Journal of the Acoustical Society
      of America, 34.

[5]   Howell P. (1990). Changes in voice level caused
      by several forms of altered feedback in fluent
      speakers and stutterers, Language and Speech,
      Vol. 33, No.4, 325-338.

[6]   Smolka E., Adamczyk B. (1983). Pomiar czestosci
      podstawowej przy oddzialywaniu echa i poglosu na
      proces mowienia jakajacych sie, Logopedia 14/15.

[7]   Webster, R.L. (1991). Manipulation of Vocal Tone:
      implications for Stuttering. In Peters, Hulstijn
      and Starkweather (Ed.). Speech motor control and
      Stuttering, pp.535-545, Elsevier Science Publishers

[8]   Bloodstein O. (1987). A Handbook on Stuttering,
      Brooklyn College of The City University of New York,
      National Easter Seal Society.

[9]   Turowski, G. (1993). Elektroniczne metody korekcji
      wad wymowy, M.Sc. thesis, Technical University of
      Gdansk, Faculty of Electronics, Department of Sound
      Engineering, Gdansk, Poland.

[10]  Piorecki P. (1994). Korektor wad wymowy oparty na
      cyfrowym procesorze sygnalowym, M.Sc. thesis,
      Technical University of Gdansk, Faculty of Electronics,
      Department of Sound Engineering, Gdansk, Poland.