_____

Record: 1

Title: 		 Talking instead of typing: Alternate access to 
computers via
speech recognition technology. 		 
Subject(s): 		 COMPUTERS & the handicapped; SPEECH 
perception;
AUTOMATIC speech recognition; HANDICAPPED -- Means of 
communication 		 
Source: 		 Focus on Autism & Other Developmental 
Disabilities ,
Summer96, Vol. 11 Issue 2, p79, 7p		 
Author(s): 		 Cavalier, Albert R.; Ferretti, Ralph 
P. 		 
Abstract: 		 Discusses the importance of alternate access 
for people
with disabilities. Ways in which speech recognition technology has been
used; Advantages of alternate access through speech recognition;
Definition of speech recognition; Efficacy of speech recognition
technology under controlled conditions.		 
AN: 		 9609101631 		 
ISSN: 		 1088-3576		 
Full Text Word Count: 		 5325		 
Database: 		 Health Source - Consumer 
Edition		 
TALKING INSTEAD OF TYPING: ALTERNATE ACCESS TO COMPUTERS VIA SPEECH
RECOGNITION TECHNOLOGY 		 



Speech recognition technology has been used extensively to enhance the
performance of persons without disabilities. In general, speech input
has proven helpful whenever optimal task performance requires the
intensive coordination of the user's hands and eyes. For many people
with disabilities, alternate access to computers through speech
recognition technology holds the promise of lessening their dependence
on others and promoting the development of their adaptive abilities. In
this article, the importance of alternate access for persons with
disabilities and the ways in which speech recognition technology has
been used to accomplish this goal are discussed. Illustrative studies of
the use of speech recognition by persons with disabilities are reviewed,
and implications for the effective application of this technology are
described. 

The popular culture is replete with representations of technological
tools that expand people's intellectual and physical capacities.
Consider, for example, the fictional Captain Kirk of the U.S.S.
Enterprise engaged in any of the following activities: requesting the
starship's computer to retrieve information about the geography of some
distant planet or the culture of some alien race, commanding the
turbolift to take him to the deck of his choice, or instructing the food
replicator to make his favorite beverage. These seemingly futuristic
capabilities are enabled by sophisticated speech recognition technology
that permits humans to control computers by simply talking to them. 

This example illustrates some of the many advantages of using speech
input to aid human-computer interaction, including the rapid retrieval
of archival information, the navigation of physical space, and more
generally, the exercise of control over the physical environment. Not
surprisingly, speech recognition technology has the potential to offer
many of the same advantages to people with disabilities, as well as
providing technological supports for helping them meet many of the
special challenges faced in school, at home, and in the workplace.
Sophisticated speech recognition technologies are increasingly available
and more affordable, raising optimism about the prospects for general
use by therapists, teachers, and clients alike. Thus, it is an
auspicious time to consider the possible benefits afforded to people
with disabilities by speech recognition technology, review
representative empirical evidence about the use of this technology by
people with various disabilities, and suggest considerations that impact
the efficacy of speech recognition in practice. 


The Advantages of Alternate Access Through Speech Recognition 


We can take it as axiomatic that the ubiquitous desktop computer is used
to increase our productivity, facilitate communication, and enhance our
personal satisfaction. Most people access their computers by means of
traditional input interfaces, such as the keyboard and the mouse.
Keyboards allow the user to compose text instructions or commands that
are then executed by the computer. Of course, the computer keyboard
represents a substantial technological advance over the goose's quill.
However, since its invention in 1870, the standard "qwerty" keyboard has
evolved in order to solve mechanical rather than human problems that no
longer exist (Norman, 1988). More intuitive and effective alternatives
(such as the Dvorak keyboard) have not been adopted because of
widespread acceptance of the "qwerty" standard. This has happened
despite the fact that the standard keyboard places a huge burden on a
novice's memory, requires serial input of pertinent information, and
necessitates the learning of a complicated syntax to effect some
operation when used for input to a computer. Some persons never even
learn the rudiments of good keyboarding. 

Direct manipulation interfaces, on the other hand, allow the user to
directly operate on iconic representations that appear on the computer's
screen through the use of a physical manipulandum, for example, using a
mouse to operate on objects in a graphic user interface. According to
Shneiderman (1982), direct manipulation interfaces have many advantages
over standard keyboards, including the direct representation of objects,
the use of physical actions on objects rather than a complicated syntax
to effect such operations, and the reversibility of operations that are
immediately visible to the user. Despite the clear advantages over
keyboarding, direct manipulation is sometimes problematic to novices who
may have difficulty mapping the mouse's movement in space onto the
computer screen. In addition, direct manipulation interfaces require
users to continuously move their hands from the keyboard to the mouse
when performing a task. "Busy hands, busy eyes" environments (Noyes &
Frankish, 1992) place considerable demands on the attentional resources
of nondisabled users, leading to less accurate and slower performance.
In general, traditional interfaces such as the keyboard and the mouse
are not transparent to the user (Norman, 1988) because they interpose
additional response requirements in order to perform a task. 

The addition of speech input capabilities to busy environments has been
associated with superior task performance by nondisabled persons
(Mountford & Garver, 1990) because the use of speech reduces the burden
on the user's attentional resources and minimizes the possibility of
response interference. For many users, speech is a relatively effortless
form of communication that can be used to select commands and functions
while they are performing manual tasks; that is, speech minimizes the
need to time share between tasks. Of course, speech can interfere with
the simultaneous performance of tasks such as listening and reading.
Therefore, care must be taken to design multitask environments that
minimize response competition or demands on common processing resources.
Nevertheless, in the general population speech input can be used to aid
performance when the user's hands and eyes are otherwise occupied. 

The proposed advantages of speech input to a computer in special
education and rehabilitation are many and will depend in large part on
the characteristics of the user, the nature of the tasks to be
performed, and the capabilities and limitations of the speech
recognition system. The most commonly cited advantages are (a) distal
input/control (i.e., no physical contact between the student/client and
the computer is necessary, except in some cases in which a microphone
wire is needed; Engelhardt, Awad, Van der Loos, Boonzaier, & Leifer,
1984); (b) flexibility in "busy hands" and "busy eyes" environments
(Noyes & Frankish, 1992); (c) reduced effort and memory load (i.e., it
can be simpler for a student/ client to speak a phrase than to key in
all of the letters or remember a control sequence; Cooper, 1987); (d)
naturalness (i.e., humans naturally or more intuitively use speech to
communicate with others and control their environment; Damper, 1984);
and (e) customization (i.e., the required inputs can be tailored to
individual students'/clients' needs; Engelhardt et al., 1984). In the
following section, the rudiments of speech recognition are discussed,
followed by illustrations of how it has been used to empower persons
with disabilities. This review focuses on representative studies in the
areas of severe physical impairment, hearing impairment, and mental
retardation. 


What Is Speech Recognition? 


With most speech recognition systems, when a user speaks into a
microphone, the computer converts the acoustic waveform (analog input)
of the vocalization into a digitized version. Spectral data are then
extracted from this digitized signal and compared to digital data sets
representing the "vocabulary items" that have been previously entered
and stored in the system. Recognition occurs when the comparison turns
up a match. The previously stored digitized items are typically called
"templates" and the user's current speech inputs are called "tokens"
(Bristow, 1986). What the computer does after a particular token is
recognized depends on the application program that is running in the
computer. Usually the application permits the user, teacher, or
therapist to have previously linked the different speech inputs to
different outputs that are tailored to the particular situation, for
example, activation of appliances in the environment, display of
information on a computer screen, or generation of synthesized speech
messages. 

Speech recognition systems can be evaluated on a number of dimensions,
including whether they are "speaker dependent" or "speaker independent,"
how easy the enrollment process is (defined below) for speaker-dependent
systems, whether they can recognize "continuous speech" or just
"isolated speech," how they are accurate with larger vocabularies, and
how well they can discriminate a user's voice in a noisy environment. 

With speaker-dependent speech recognition systems (the most common type
available to teachers and therapists), a user must create the templates
for all of the vocabulary items that are needed to operate a speech
recognition application prior to actual use of the application. This is
typically done by having the user vocalize 3 to 10 samples of each
vocabulary item into the system--a process referred to as "enrollment."
Depending on the system, the template of a vocabulary item is composed
of the digital data either for each separate sample or for the average
of all samples. Because of this pattern-matching approach, the more
consistently a user vocalizes vocabulary items, the better a system's
recognition rate. A user, teacher, or therapist typically can adjust a
system's "recognition threshold," that determines how closely a token
must match a template in order for a recognition to occur (Bristow,
1986). Speaker-independent systems, which are typically far more
expensive, require little or no enrollment and consequently have
particular value in situations in which a large number of people need to
interact with the same system. 

"Continuous speech" systems can recognize fluid conversational speech
and word-processing-like dictation. In contrast, "isolated speech"
systems (the more commonly available ones) require a user to pause
between vocabulary items. "Items" are not necessarily restricted to
single words, but rather can include phrases consisting of a few words
(e.g., "Open the door."). The controlling factor is typically the
duration of the utterance, with most isolated speech systems allowing
vocalizations from 1 to 3 seconds in duration. 


Educational and Clinical Research 


In general, the range of potential educational and rehabilitative
applications of speech recognition technology is large. Researchers have
begun to explore these applications and advance our understanding of
some of the factors necessary to optimize their benefits. In the
following section, a review of some of the research literature is
provided to illustrate ways in which speech recognition has been used to
assist people with disabilities. Research articles were selected on the
basis of two considerations: (a) The study evaluated the efficacy of
speech recognition technology under controlled conditions (i.e., a
research design was used to evaluate the effects of the technology); and
(b) the studies illustrate a range of applications across the categories
of physical impairment, hearing impairment, and mental retardation. 

Persons with Physical Disabilities. Fried-Oken (1985) examined speech
recognition with a 10-year-old boy who had quadriplegia and used a
respirator and with a 19-year-old young man who had spinal cord and
brain injury. The child used 23 different voice commands to operate a
mathematics software program in an Apple II-based system. Due to
variation in his pronunciations and interfering noise from his
respirator, the 10-year-old had recognition rates that ranged from 40%
to 60%. Despite these relatively low rates, he expressed increased
motivation to continue learning and to continue using the speech
recognition system. The young man, using the aviation code for the
alphabet (e.g., alpha, beta, charlie), had recognition rates ranging
from 79% to 96%. Once it was established that the system could recognize
the young man's dysarthric speech, he began to work on producing minimal
pairs (a set of words that differ by one sound) for recognition. Correct
articulation of the minimal pairs was associated with the opportunity to
play a computer game. During this phase of training, about 83% of all
tokens were recognized. Fried-Oken (1985) claimed that the latter
exercise improved the young man's articulation. However, information
about the clarity of his articulation before speech recognition training
was not provided. 

Treviranus, Shein, Haataja, Parnes, and Milner (1991) investigated
whether the combination of speech recognition and the scanning method of
selection would improve the input rate and accuracy in using
augmentative communication systems for people who were severely
physically disabled. Prior to the study, the 6 participants used
discrete switches with scanning to access their computers. Although they
were nonspeakers, each could make three or more repeatable (but not
necessarily intelligible) sounds. Treviranus et al. reported only the
results for a 12-year-old boy with cerebral palsy who had three
distinguishable vocal utterances. He used his Macintosh-based
communication system for both writing and "speaking," and controlled the
scanning mode in row-column fashion with two head switches. During
formal testing on writing tasks, the boy averaged approximately 2.35
selections per minute using his traditional input method and 3.14
selections per minute using speech recognition and scanning. The average
total errors were 3.6 and 3.9, respectively. The researchers noted that
speech recognition was far more adversely affected by stable
environmental noise (e.g., an overhead fan or heating unit) than by
other classroom noises (e.g., voices of classmates or chairs being
moved). Data for the other participants were not reported by the
authors. 

Schmitt and Tobias (1986) configured an Apple II-based system with
speech input linked to synthesized speech and environmental control
output for an 18-year-old young woman who was severely physically
disabled and visually impaired. Her physical disability prevented her
from using the keyboard and any multiple switch system; her visual
impairment prevented her from reading anything on the display screen.
After enrollment, the vocabulary was divided into separate "pages," or
subsets of related items, so that only a portion of the total vocabulary
was available for recognition at any one time. This was done to improve
the recognition accuracy of the system. The young woman could "flip"
pages under speech command to activate the subset containing the
vocabulary item of her choice. Under these conditions, she achieved a
recognition rate of 54%. This low rate was due in part to her low
frustration tolerance: When she became frustrated, she experienced
heightened emotion, which in turn altered her pronunciations. The
researchers also reported that enrollment was a tedious process for her.
They stated anecdotally that an unanticipated positive result of system
use was improvement in the intelligibility of her natural speech. 

Coleman and Meyers (1991) compared the speech recognition of an Apple
II-based system on vocalizations of 10 adults with cerebral palsy and
dysarthria and 13 adults without disabilities. Prior to the study, the
single-word intelligibility of the adults with dysarthria averaged about
57%, and their word intelligibility in sentences averaged about 51%. The
researchers sought to determine if the type of language item (i.e.,
vowels, consonants, easy-to-pronounce words, and hard-to-pronounce
words) influenced recognition accuracy with either of these two
populations. Across all language types, the system recognized
considerably fewer tokens produced by the persons with cerebral palsy
than by the persons without disabilities. However, the pattern of
recognition errors between language types was remarkably similar. In
general, consonants were recognized less frequently than were the three
other types. 

For people with physical disabilities, speech recognition technology
holds the promise of increased environmental control, freedom of
movement, and access to a wider range of personal choices in all
settings. The studies reviewed in this section show that the addition of
speech input can increase the rate of choices made by people with
physical disabilities despite the fact that speech recognition
performance is error prone. In addition, at least one anecdotal report
indicated that the use of speech recognition technology is associated
with improvements in clients' motivation and the intelligibility of
their speech (Fried-Oken, 1985). Some possible drawbacks to the use of
this technology have also been noted, including a client's frustration
with system errors (Schmitt & Tobias, 1986), fatigue from producing
speech (Schmitt & Tobias, 1986), and inaccurate recognition as a result
of ambient background noise (Treviranus et al., 1991). Surprisingly, the
pattern of recognition errors seems to be quite similar for dysarthric
and normal speech. The latter finding holds out the hope that general
improvements in speech recognition technology may directly benefit
persons with physical impairments. 

Persons with Hearing Impairments. Stevens and Bernstein (1985) compared
the accuracy of human listeners who did not have disabilities with that
of speech recognition systems in recognizing the speech of 5 persons who
were deaf. The speech of persons who are deaf is especially challenging
for speech recognition technology because of frequent within word and
between-word pauses. The human listeners recognized 7% to 61% of the
words compared with the speech recognition system's rates of 75% to 99%.
Similar results were reported by Carlson and Bernstein (1987). 

Abdelhamied, Waldron, and Fox (1990) developed a two-step segmentation
method to increase the accuracy of speech recognition for speakers with
heating impairments. In essence, this method first estimates the
inter-and intrabreak pauses of the speaker who is deaf, divides the
speech stream into possible speech segments, and then compares all
possible groupings of consecutive speech segments against reference
templates. The recognition accuracy of this system was then determined
for 2 deaf and normally hearing male speakers who spoke both isolated
and connected words. The system's accuracy in recognizing isolated words
averaged 93% and 98% for persons who were deaf and for those without
hearing impairments, respectively. The system was less accurate in
recognizing connected words, averaging 82% and 86% for persons who were
deaf and for those without hearing impairments. In general, errors of
recognition were attributable to the similarity of words, the difficulty
in segmenting the speech stream, and the variability in the person's
speech. 

Taken together, these studies show that methods are available for
enhancing the accuracy of speech recognition systems for persons with
hearing impairments and that these systems may show greater recognition
accuracy than do listeners without hearing impairments. This latter
finding may be especially important to teachers and therapists, who may
have considerable difficulty understanding the speech of persons with
hearing impairments. Finally, Stevens and Bernstein's (1985) finding
that speech recognition technology may be able to interpret dysarthric
speech better than humans offers exciting possibilities for improved
communication for children and adults with speech impairments, for
example, through the automatic translation of their natural speech into
intelligible digitized or synthesized speech output. 

Persons with Mental Retardation. Brown and Cavalier (1992) conducted
research with a woman with severe mental retardation and quadriplegia
who had been institutionalized for most of her life. The woman showed
evidence of learned helplessness by virtue of being dependent on other
persons for most of her daily needs throughout her life. Using an Apple
II-based system, they investigated whether she could learn that her
voice was now the controlling agent for devices in her environment and
that different vocalizations controlled different devices. The study
took place in the general activity area of her dormitory, the noise
level of which probably contributed to the need to frequently repeat the
enrollment process in order to maintain an acceptable level of
recognition accuracy. Results indicated that she effectively learned and
differentiated the cause-and-effect relationships despite a recognition
rate of only 59%. She also demonstrated significant positive emotions
and became more "animated" when she was in control of her environment. 

To reduce the cognitive load and facilitate users' choice making, Brown,
Cavalier, Sauer, and Wyatt (1992) developed an IBM-based system that
displays photographic-quality color images representing both the
environmental devices and the digitized speech messages that a user can
speech-activate. The system was tested with a girl of 11 years of age
and a woman of 21 years of age, both of whom were severely mentally
retarded and physically disabled. Choices could be made through direct
selection by voice, linear scanning by voice, or row-column scanning by
voice. Both participants were reported to have learned the associations
among the vocalizations, the graphic images, and the resultant
activations. However, data about these performance changes were not
included in the report. Finally, the performance of 1 client was said to
have deteriorated after the introduction of a third choice. In the Brown
and Cavalier (1992) and the Brown et al. (1992) studies, the performance
of both speech recognition systems was significantly affected by the
placement of the microphone. 

These studies suggest that persons with severe and profound mental
retardation can learn to use speech recognition technology to control
choices and can discriminate the relationships among different
vocalizations and their associated activations. As with other studies,
ambient noise reduced the accuracy of speech recognition and at least 1
client seemed to lose interest in using the system. Nevertheless, these
studies seem to show that this technology enables persons with severe
and multiple disabilities to use their natural speech to communicate and
control their environment. Perhaps the use of this technology can lessen
their dependence on others and break the cycle of learned helplessness
that often characterizes persons with these conditions. 


Methodological Analysis of the Knowledge Base 


Most of the studies reviewed here provide some quantifiable information
about the accuracy of speech recognition systems used by persons with
disabilities. For the most part, these systems proved to be relatively
robust in the face of the nearly unintelligible speech of some clients.
In fact, speech recognition systems have sometimes been shown to be more
accurate than normal human listeners. 

Researchers often provide anecdotal reports about the benefits of speech
recognition technology (Fried-Oken, 1985; Schmitt 8: Tobias, 1986;
Treviranus et al., 1991). Some researchers claim that the use of speech
recognition technology improves the intelligibility of clients' speech,
their affective state, or their motivation for learning (Brown &
Cavalier, 1992; Fried-Oken, 1985). Of course, these reports increase our
optimism about the benefits of speech recognition technology. However,
these claims should be viewed with some skepticism because solid
empirical information about these purported benefits is not provided.
For example, it is possible to measure changes in the intelligibility of
human speech before and after training with speech recognition
technology. Changes in intelligibility as a result of training could
then be compared to changes that occurred for people who did not use
this technology. If experience with speech recognition technology
increases the intelligibility of human speech, then children who use
this technology should improve more than those who did not use the
technology. Studies like this would enable professionals to document the
genuine benefits to persons with disabilities of using speech
recognition technology. 

Many of the studies reviewed here provide information about the average
level of recognition accuracy after extended use of the speech
recognition system. Information about the average level of accuracy is
potentially misleading, however, because it may mask changes in
performance that are likely to occur over time and with practice.
Detailed information about these changes would provide information about
factors that affect the use of these systems, such as learning, fatigue,
frustration, and the like. Therefore, progress in assessing the
usefulness of speech recognition technology depends on careful
measurement of changes in performance over time and under different
conditions. 

Some might argue that it is difficult to measure these kinds of factors
and that the highly idiosyncratic nature of many disabilities makes
group comparisons impossible. Although sensitivity to these concerns is
important, it is equally important to recognize that resourceful
researchers have many measurement and research design strategies that
can be used to assess individual clients. Unfortunately, the
contemporary instructional and therapeutic landscape is littered with
ineffective treatments that were once eagerly embraced by hopeful
parents and teachers. Carefully controlled research will enable
professionals to better assess the validity of claims made about the
benefits of speech recognition technology. 


Recommendations 


The present review of research studies and an analysis of a variety of
other clinical reports permit recommendations about strategies for
optimizing the use of speech recognition technology in practice. These
recommendations may be especially useful to teachers and therapists who
are seriously considering the possibility of using speech recognition as
an alternate access modality. They take into account the enrollment
process, user control and feedback, and the system's vocabulary. 

Enrollment Process. The enrollment process is crucial to the accuracy
and reliability of the speech recognition system. During the enrollment
process, there can be no mistakes or sloppiness in the creation of the
voice templates. Recognition rates for all of the templates could be
affected by carelessness in the creation of one (Brown & Cavalier, 1992;
Poock, 1991). Further, it is best to conduct the enrollment process
under the typically noisy environmental conditions in which the speech
recognition application will eventually be used (Damper, 1984; Noyes &
Frankish, 1992). Finally, enrollment should be checked anti reconducted
at periodic intervals. The human voice changes over time, under stress,
and with physical exertion (Brown & Cavalier, 1992; Cooper, 1987).
Failure to periodically monitor enrollment integrity will result in
lower levels of recognition accuracy in practice. 

User Control and Feedback. Even the best speech recognition system will
err, and some errors could have serious repercussions for the user.
Therefore, the system should be configured so that the user is always
provided feedback about the recognizer's decision regarding the last
input. This is important because users typically lack confidence in
their input because there is no inherent tactile or auditory feedback
from the recognizer (Damper, 1984; Treviranus et al., 1991). Related to
this, users should be provided with a means of confirming that the
system's recognition is indeed the intended recognition. Error recovery
facilities should be provided so that the user can avoid activation of
an unwanted environmental device or speech output (Cooper, 1987; Noyes &
Frankish, 1992). 

System Vocabulary. Four characteristics of the system's vocabulary have
been shown to affect the accuracy of speech recognition. First, it is
best to restrict the number of active vocabulary items as much possible
without unnecessarily constraining the user's range of choices.
Recognition rates typically decrease as the number of voice templates to
be compared increases (Engelhardt et al., 1984; Noyes & Frankish, 1992).
Second, vocabulary items that are as phonetically dissimilar as possible
should be chosen. Recognition rates typically decrease as the similarity
among the stored templates increases (Damper, 1984; Schmitt & Tobias,
1986). Finally, it is generally best to choose vocabulary items that are
long (i.e., multisyllabic) rather than short, and items that are common
rather than rare, whenever possible. Longer items permit more spectral
data to be extracted, and common items are typically pronounced more
consistently (Damper, 1984; Fried-Oken, 1985). Lastly, whenever
possible, users should be encouraged to vocalize authoritatively during
use of speech recognition technology. Recognition rates usually will be
improved (Poock, 1991). 


Conclusions 


Children with disabilities often experience a world that is so poorly
matched to their abilities that they are effectively excluded from
opportunities to promote their social, emotional, and intellectual
development. Although computers are the most powerful and versatile
tools available to these children, alternate access to these potentially
powerful systems has typically been slow and restrictive, thereby
limiting the potential benefits to prospective users. As a consequence,
children with disabilities fall further and further behind their
nondisabled peers. This point becomes especially salient as the
initiative to educate students in inclusive settings gains momentum
(Treviranus et al., 1991). 

The reports summarized above demonstrate how speech recognition
technology can be used to help improve a student's communication, active
control of the environment, and learning. This technology also offers
exciting possibilities for directly advancing curricular objectives. For
example, because people typically speak faster than they write and
beginning writers typically find writing by pencil (or even word
processor) to be arduous, the automatic recognition and transcription of
their oral language might reveal a more coherent representation of the
student's thoughts (Wetzel, 1991). In addition, students may find the
experience to be more natural and less frustrating. Further, imagine the
possibility that a computer is capable of identifying (and possibly
correcting) syntax errors. A system such as this would enable a student
with disabilities to draft and then revise his or her writing. In
similar fashion, a teacher could also incorporate the system into the
student's reading instruction by having the student read his or her own
written compositions. These and many other creative applications may
someday tap the instructional power of speech recognition technology. 

In an increasingly electronic and computer-dependent culture, speech
recognition technology has the potential to provide many persons with
disabilities improved and more equitable access to education,
communication, and employment. Although this potential has only begun to
be realized, it is possible to envision the day when persons with and
without disabilities interact with their tools as naturally and
effectively as an imaginary starship captain in a fanciful television
series. 


AUTHORS' NOTE 


The preparation of this manuscript was supported in part by Grant No.
H180E30043 from the U.S. Department of Education, Office of Special
Education Programs, Division of Innovation and Development, to the
second author. 


REFERENCES 


Abdelhamied, K., Waldron, M., & Fox, R. A. (1990). Automatic recognition
of deaf speech. The Volta Review, 92, 121-130. 

Bristow, G. (Ed.). (1986). Electronic speech recognition: Techniques,
technology and applications. New York: McGraw-Hill. 

Brown, C. C., & Cavalier, A. R. (1992). Voice recognition technology and
persons with severe mental retardation and severe physical impairment:
Learning, response differentiation, and affect. Journal of Special
Education Technology, 11, 196-206. 

Brown, C. C., Cavalier, A. R., Sauer, M., & Wyatt, C. (1992). The
sound-to-speech translation system utilizing photographic-quality
graphic symbols. The APL Technical Digest, 13, 482-489. 

Carlson, G. S., & Bernstein, J. (1987). Speech recognition of impaired
speech. In R.D. Steele & W. Gerrey (Eds.), Proceedings of the Tenth
Annual Conference on Rehabilitation Technology (pp. 103-105).
Washington, DC: RESNA. 

Coleman, C. L., & Meyers, L. S. (1991). Computer recognition of the
speech of adults with cerebral palsy and dysarthria. Augmentative and
Alternative Communication, 7, 34-42. 

Cooper, M. (1987,March/April). Human factor aspects of voice
input/output. Speech Technology, pp. 82-86. 

Damper, R. I. (1984). Voice-input aids for the physically handicapped.
International Journal of Man-Machine Studies, 21, 541-553. 

Engelhardt, K. G., Awad, R. E., Van der Loos, H. F. M., Boonzaier, D.
A., & Leifer, L. J. (1984). Interactive evaluation of voice control for
a robotic aid: Implications for training and applications. Proceedings
of the Voice I/O Systems Applications Conference. Arlington, VA:
American Voice Input/Output Society. 

Fried-Oken, M. (1985). Voice recognition device as a computer interface
for motor and speech impaired people. Archives of Physical Medicine and
Rehabilitation, 66, 678-681. 

Mountford, S. J., & Gaver, W. W. (1990). Talking and listening to
computers. In B. Laurel (Ed.), The art of human-computer interface
design (pp. 319-334). Reading, MA: Addison-Wesley. 

Norman, D. (1988). The psychology of everyday things. New York: Basic
Books. 

Noyes, J. M., & Frankish, C. R. (1992). Speech recognition technology
for individuals with disabilities. Augmentative and Alternative
Communication, 8, 297-303. 

Poock, G. K. (1991). Fifteen steps toward success: Knowing the
personality of your speech recognizer. Unpublished manuscript, Naval
Post Graduate School, Monterey, CA. 

Schmitt, D. G., & Tobias, J. (1986). Enhanced communication for a
severely dysarthric individual using voice recognition and speech
synthesis. In M. Donath, H. Friedman, & M. Carlson (Eds.), Proceedings
of the Ninth Annual Conference on Rehabilitation Technology (pp.
304-306). Washington, DC: RESNA. 

Shneiderman, B. (1982). The future of interactive systems and the
emergence of direct manipulation. Behavior and Information Technology,
1, 237-256. 

Stevens, G., & Bernstein, J. (1985). Intelligibility and machine
recognition of deaf speech. In C. Brubaker (Ed.), Proceedings of the
Eighth Annual Conference on Rehabilitation Technology (pp. 308-310).
Washington, DC: RESNA. 

Treviranus, J., Shein, F., Haataja, S., Parnes, P., & Milner, M. (1991).
Speech recognition to enhance computer access for children and young
adults who are functionally nonspeaking. In J. J. Presperin (Ed.),
Proceedings of the Fourteenth Annual Conference on Rehabilitation and
Assistive Technologies (pp. 308-310). Washington, DC: RESNA Press. 

Wetzel, K. (1991). Speech technology II: Future software and hardware
predictions. The Computing Teacher, 19, 19-21. 

~~~~~~~~

By Albert R. Cavalier and Ralph P. Ferretti 

Albert R. Cavalier, PhD, is an associate professor in the Department of
Educational Studies and director of the Center for Assistive and
Instructional Technology at the University of Delaware. His research
interests are in human-factors principles involved in the use of
assistive technology, and the cognitive demands placed on students with
disabilities by functional tasks and assistive devices. Ralph P.
Ferretti, PhD, is an associate professor in the Departments of
Educational Studies and Psychology at the University of Delaware. His
research interests focus on cognitive mechanisms that affect the use of
assistive technology for persons with learning handicaps. Address:
Albert R. Cavalier, Department of Educational Studies, University of
Delaware, Newark, DE 19716. 

  _____  

Copyright of Focus on Autism & Other Developmental Disabilities is the
property of PRO-ED and its content may not be copied or emailed to
multiple sites or posted to a listserv without the copyright holder's
express written permission. However, users may print, download, or email
articles for individual use. 
Source: Focus on Autism & Other Developmental Disabilities, Summer96,
Vol. 11 Issue 2, p79, 7p
Item: 9609101631
  _____



This e-mail was generated by a user of EBSCOhost who gained access via
the EDISON HIGH SCHOOL account. Neither EBSCO nor EDISON HIGH SCHOOL are
responsible for the content of this e-mail.