Inclusive and Supportive Education Congress International Special Education Conference Inclusion: Celebrating Diversity? 1st - 4th August 2005. Glasgow, Scotland

home about the conference programme registration accommodation contact

Speech Recognition for Students with Disabilities

Paul D. Nisbet, Allan Wilson, Dr. Stuart Aitken
Communication Aids for Language and Learning (CALL) Centre,
Moray House School of Education
 University of Edinburgh, Edinburgh, Scotland
Paul.Nisbet@ed.ac.uk

Abstract

Speech recognition software has potential to offer a means of writing and recording to students with literacy or visual difficulties. However, while some students have successfully used speech recognition, others have not found it effective. This paper presents results of a project that investigated best practice in schools where speech recognition was being used successfully, and then developed and evaluated training materials to help staff and students use speech recognition productively. 40 mainstream secondary schools in Scotland participated in the project. Each school was provided with Dragon Naturally Speaking or IBM ViaVoice software, technical support and training delivered on site, plus a training pack designed to be used with one or other program. 23 schools returned evaluations, detailing results for 32 students. Results showed that the training packs were well received and effective. 70% of the students intended to continue using speech recognition after going through the training; 3% were not sure; and 25% did not intend to continue using speech recognition. The most common reasons for not continuing to use speech recognition in school were: other ICT tools were judged to be more suitable; timetable/lack of access; and unsatisfactory accuracy. Training and support was probably the most significant factor to influence success: 91% of the students who intended to continue using speech recognition received training once or more per week. The most common reasons for continuing to use speech recognition were speed, legibility and accuracy of spelling, and independence. Staff did not report any significant difference in effectiveness between Dragon NaturallySpeaking and IBM ViaVoice. Staff evaluated student skills before and after using speech recognition and in most areas there was improvement. The results showed that speech recognition software can provide an effective method of recording for students with disabilities. The main factors for successful implementation were frequency of training sessions, and to a lesser extent, student aptitude.

Background

Many students with additional support needs have difficulty with writing and recording, due to motor difficulties, visual impairment, specific learning difficulty, or other learning difficulty.   Speech recognition software (where the student dictates into a microphone and the program converts the speech into written text) has been shown to provide an effective method of writing for some students with additional support needs (O’Hare & McTear, 1999; Becta, 2000; Donegan, 2000)

In recent years, the accuracy and ease-of-use of speech recognition software has improved considerably, while the cost has fallen.   As a consequence of this and the research mentioned above, many pupils, parents and schools have purchased speech recognition systems in order to try and overcome pupils’ difficulties with writing or recording.   In some cases, speech recognition has had a considerable impact and has enabled a student with additional support needs to write and create longer and more complex pieces of work, more independently (Donald, 1998; Litten, 2000) However, through discussion with suppliers, practitioners, and also through feedback from staff, patients and students who have made use of the CALL Centre Assessment and Information services, there are also many students and schools were known by the project team to have explored speech recognition but have not found that it was not effective.

Research Questions

The primary purpose of this project was to investigate best practice in schools where speech recognition was being used successfully, and develop and evaluate training materials based on the results of the investigation, in order to help staff and students use speech recognition more successfully.   In addition, we wished to investigate supplementary questions concerning the:

• effectiveness of two different speech recognition programs;
• influence of student skills on successful use of the software;
• effect of use of speech recognition on basic literacy.  

The focus of the project was on mainstream secondary schools partly because we suggest that the largest group of students potentially to potentially benefit from speech recognition are those with specific learning difficulties in secondary schools. This is because: speech recognition software is primarily designed for adult, business users and voices (and so is less effective with younger children or those with speech or language difficulties); the quantity of writing which has to be completed independently increases considerably in secondary school; and the process of inclusion into mainstream in Scotland has resulted in an increase in the number of children with additional support needs who experience difficulties with writing. For example, the number of students using special arrangements (such as scribe, reader, extra time, transcription of the paper) in Scottish Qualification Authority examinations increased from 3,094 in 1995 to 9,904 in 2004 (SQA, 2005).

Method

The CALL Centre is an action research and service organisation. The Centre provides assessment support to children and schools across Scotland; an information and publications service; a programme of training and staff development; loan of equipment for evaluation; and undertakes research and development (CALL Centre, 2004). The aim of this project was primarily to develop and evaluate a training resource for staff, and the research design drew upon our experience working directly in schools. We adopted a methodology that involved minimal input from project staff, that deliberately targeted a range of schools and local authorities across Scotland, and that was relatively ‘loose’, in order to gather evidence of effectiveness of the intervention in ‘real’ settings.

• The aim of this project was primarily to develop and evaluate a training resource for staff, and the research design drew upon the project team’s experience of working directly in schools. We adopted a methodology that deliberately targeted a range of schools and local authorities across Scotland, included pupils with a range of support needs and required minimal input from project staff. These control measures were introduced for three reasons. First, we included several local authorities to reduce the possibility that any influence on results would be specific to a single authority, possibly reflecting specific support structures. Second, we included a range of schools and pupil support needs to ensure that results were not specific to one type of support need in one type of setting. The final control introduced, to reduce input from project team members, was a direct attempt to reduce observer bias. Taken together, results obtained with these controls in place would help to ensure that findings could more accurately be translated into effective school practice: the results would have increased validity and reliability.

First drafts of two training books were written, one for each of two speech recognition programs (IBM ViaVoice Millenium Pro 7 and Dragon NaturallySpeaking Preferred 4). The books ran to 100 pages and included:

• a question and answer section designed to help staff assess students who would be likely to benefit from use of speech recognition;
• a self-study tutorial for staff;
• ten lesson plans for training students (following from the SNOW Dragon Dictate training (Lubert & Campbell, 1998).

While the training resources were being researched and written, we contacted schools, ICT specialists and the authorities in Scotland to identify schools to take part in the project. Forty schools in eight different local authorities across Scotland expressed an interest. The procedure for implementing the project in each school was as follows:

• Schools identified a lead member of staff to co-ordinate the project, and one computer capable of running the speech recognition program in a suitable location (for example, a Learning Support Base or other area which was readily accessible to students).
• Project staff delivered a half-day staff development session, to install the speech recognition software, train staff and discuss which students might be involved in the project. Half the schools were provided with IBM ViaVoice Millenium Pro 7, while the other half received Dragon NaturallySpeaking Preferred 4 or 5. Schools were also given the appropriate CALL Centre Training pack for the software, and a high quality microphone.
• The school identified up to three students, who in their opinion would benefit from using speech recognition, to participate in the project. (Some schools worked with a larger number of pupils but we advised that staff should start with a small number initially.) Students who were, in the judgment of staff, unlikely to benefit from using speech recognition , in the judgment of staff, were not included in the study.
• Following the first training session, teachers reviewed the training pack and practised exercises in each lesson the to become confident with the speech recognition software and the training pack.
• Once staff had become familiar with the program and the training pack there was an option to contact the project team for a second half-day of training. On the return visit, the team supported teachers through the first lesson with the student and attempted to resolve any problems or issues that had arisen with the program or its use in school.
• Thereafter, staff worked through the training pack with the pupils that they had identified. The precise number and format of teaching sessions with pupils was determined by the staff, although we recommended that if possible, teachers would work with the pupils on a one-to-one basis, or in small groups, at least for the first three or four speech recognition sessions.   We advised that training sessions should be timetabled to take place at least once or twice per week. Staff were able to contact the project team by telephone and e-mail for support if necessary.

Staff development in the 40 schools took place over 9 months, from October 2000 to June 2001, followed by a further 7- month period for schools to use and evaluate the speech recognition software. As the project progressed, comments and suggestions from staff were received and incorporated to revised versions of the training pack. The resources were updated as new versions of the software were released: the final books were written for ViaVoice Pro 9 (Nisbet & Wilson, 2002a) and NaturallySpeaking Preferred 5 (Nisbet & Wilson, 2002b).

Staff were asked to complete three data collection forms:

• Pupil Profile, recording information about each student involved in the project;
• School Record, with background details about the school and the computers used;
• Pupil Evaluation Questionnaire, to be completed by each student, in collaboration with teachers, to gather views and comment on the Training Pack and the speech recognition software.  

Results

Responses

Of the 40 schools that received training and speech recognition software and hardware, 23 (57.5%) schools returned Evaluation Questionnaires in respect of 32 pupils. Subsequent follow-up with the staff from the seventeen schools who did not return questionnaires revealed the following explanations for not proceeding with the project to introduce speech recognition:

• seven schools did not give any particular reason
• five schools cited staff changes or absences;
• four schools stated lack of time and other priorities;
• four schools reported technical problems, lack of access to suitable computers, and in one case the designated laptop was stolen;
• one school noted that the pupil had found keyboarding more effective;
• one school reported that the pupil who had been identified to take part in the project had left the school.

Effectiveness of the training package

We attempted to measure the effectiveness of the training package by asking students if they intended to continue using speech recognition having gone through the training. Of the 32 students in 23 schools:

• 23 (72%) students reported they would continue using speech recognition;
• 1 (3%) was not sure;
• 8 (25%) did not intend to continue using speech recognition.

Evaluation of the training pack

Feedback from staff about the training provided by the project team, and the training resource, indicate that both were well received and perceived to be useful. 20 (69%) staff regarded the training pack as ‘excellent’; 8 as (28%) ‘useful’ and 1 (3%) ‘not needed’. Initial training was regarded as even more important than the Pack (23 staff rated it as ‘excellent’), illustrating the importance of good training and support delivered on-site.

Figure 1:   Feedback on CALL training, pack and support

Reasons for not continuing to use speech recognition

The most common reason for not continuing to use speech recognition (given by six of the eight students) was that, having evaluating it, they decided that other software or technology, such as keyboarding or word prediction, was more effective. Some commented on poor accuracy of the speech recognition, frustration, lack of opportunity for practice, articulation difficulties, and unsatisfactory accommodation for dictating. A change of timetable prevented one student from completing the training while another specifically reported that the reason for lack of success was due to poor accuracy and reliability.

Reasons for continuing to use speech recognition

The most common reasons given by students for continuing to use speech recognition were:

• to enable them to tackle extended writing tasks more successfully (reported by 12 students);
• the speed at which text could be produced (5 pupils);
• legibility (2 students);
• accuracy of spelling (4 students);
• independence (1 student).

Students involved in the project

Evaluation questionnaires were returned with respect to 32 pupils: 26 male, 5 female and one was not recorded. The majority of the students (11) were in second year of secondary schooling, those in third and fourth year numbered seven each, five students were in first year, and two students in fifth year. For the majority (27 out of 32) of students the description given for participation was dyslexia; four were noted as dyspraxia; four had difficulty with handwriting due to muscular atrophy, arthritis or cerebral palsy. One student had severe physical difficulties due to arthogryphosis.

The Pupil Profile records completed by staff provided an indication of pupil abilities across a range of skills, scored with respect to an ‘average’ pupil. Figure 2 provides the mean scores for the students, grouped according to those who will, and will not, continue to use speech recognition. Given the small number of subjects the results are not statistically significant but do raise some interesting issues.

• Figure 2:   Mean student abilities, scored by staff

(Based on scores for 17 pupils who intend to continue with speech recognition and 3 who do not)

• Writing and spelling skills of all the students were described as poorer than average: the main reason, presumably, for them participating in the project.
• Staff judged the students’ reading skills to be poorer than average.
• All students were reported as having above-average perseverance, ability to work independently and motivation to use speech recognition, with little difference between those who were and were not successful with speech recognition. Students’ motivation to use speech recognition was either good or excellent. This reflects the advice from CALL, based on a review of research, which suggested that students with strong motivation, perseverance, and ability to work independently are more likely to be successful with speech recognition.
• To use speech recognition effectively, a student must be able to plan, compose, and articulate satisfactory English text.   Students who have difficulty in any of these areas are likely to have difficulty using speech recognition effectively. In eight out of thirty-two cases, staff specifically commented that students’ accent or quality of speech had an adverse impact upon accuracy of recognition.   Poor accuracy was a key factor for five of the students deciding not to continue using speech recognition.

Effect of speech recognition on student skills

The project was not designed to rigorously investigate rigorously the effect of speech recognition on student skills. Given that results from other studies (Higgins &   Raskind, 2000; Miles et al, 1998; Sanderson & Smits, 2001) which have reported improvements in reading and spelling skills attributed to using speech recognition, we asked staff to estimate changes in student skills across a number of areas.   Figure 3 presents the mean changes in abilities reported by staff. We also asked staff to test literacy skills using standardised measurements before and after going through the speech recognition training.   Only four staff undertook this testing, with five students, and results were inconclusive.   Some students achieved significant improvement (for example, an improvement of 3.6 years in sentence comprehension, over a four-month period), while with others there was no apparent effect.

Figure 3: Mean changes in abilities reported by staff

  (Based on 17 pupils who intend to continue with speech recognition and 6 who do not)

Effect of frequency of use

Staff were asked to record the frequency of students’ speech recognition sessions and the results demonstrate clearly that training and practice is essential to achieve success using speech recognition software.   91% of students who intended to continue using speech recognition received training once or more per week, compared to 37% of students who did not intend to continue using the software.  

Figure 4:   Frequency of practice of students who will and will not continue with speech recognition

(Based on 23 students who will continue to use SR and 8 who will not)

Hardware and software issues

Research and case studies, together with comments from suppliers and staff who have used speech recognition in schools, underline the importance of using appropriate technology, and so we asked teachers to provide technical details of the computers that they were using to run the speech recognition programs.   We also asked staff to rate the following aspects of the speech recognition program, on a scale from one to five:

• reliability;
• ease of enrolment (the process of initial training for each user’s voice);
• accuracy;
• method of correction (of miss-recognised words);
• overall.

The technical characteristics of the computer used seemed to have little impact on the outcome of scores given for each of the above factors: for example, staff in one school who used a Pentium II, Windows 95 computer with 32 MB of RAM gave the program an overall score of 4 out of 5, while software installed on a computer with a Pentium III processor and 128 MB of RAM was rated 1 out of 5.   The majority of the computers used were Windows 98 machines, with Celeron processors, running at between 300 and 600 MHz, with 64 MB of RAM. (The recommended specification for running NaturallySpeaking version 5 was a Pentium 266 with 64 MB of RAM; ViaVoice 7 required a Pentium II PC with 48 MB.)

The scores reveal little difference in effectiveness between NaturallySpeaking and ViaVoice systems, with mean overall scores of 3.38 and 3.42 respectively.  

A consideration of the ‘success rate’ for each program (i.e. the number of children who reported they would continue to use the software) suggests that ViaVoice was more effective than NaturallySpeaking (12 out of 14 students said they would continue to use it, compared with 11 out of 17 for NaturallySpeaking), but it is more than likely that other factors such as staff and student skills, computer specifications and especially, frequency of practice had greater effect.

Figure 5:   Mean ratings, from staff, of the speech recognition programs

(NaturallySpeaking was scored by 16 staff, ViaVoice by 12)

Conclusions

The study demonstrated that speech recognition software can provide a practical writing method for some students with additional support needs. Training for staff and students is essential for successful uptake and the training package developed during the project was well received by staff and effective in helping teachers to train students to use speech recognition. The most common reasons given by students for continuing to use speech recognition, having gone through the training process, were, improved writing speed, legibility, spelling accuracy and independence. The most common reasons for not continuing to use speech recognition were: other writing tools were more effective; lack of access or opportunity; and unsatisfactory accuracy. For the majority of students, staff estimated that student skills, across a number of the areas, had improved over the course of using speech recognition.

Acknowledgements

This work was funded by the Scottish Executive Education Department. We are grateful to teachers and students in all the schools who participated in the project, and to staff at iANSYST and Words Worldwide for technical support.

References

BECTa (2000) SEN Speech Recognition Project Final Report. BECTa. http://www.BECTa.org.uk/technology/speechrecog/project/index.html

CALL Centre (2004) CALL Centre Annual Report 2004-2005 . CALL Centre, University of Edinburgh. www.callcentrescotland.org

Donald, E. (1998) Voicetext: Evaluation of System when used by Students with Special Educational Needs. Perth High School, Perth, Scotland (unpublished)

Donegan, M. (2000) Voice Recognition Technology in Education: Factors for Success Ace Centre Oxford. ISBN 1 903303 00 1.

Higgins, E. L. & Raskind, M. H. (2000). Speaking to Read: The Effects of Continuous vs. Discrete Speech Recognition Systems on the Reading and Spelling of Children with Learning Disabilities. Journal of Special Education Technology, 15 (1), 19-30.

Litten, M. (2000) Usable Voice Recognition Software: The Mark College Experience! Dyslexia Review, Vol. 13 No. 2. www.keyspell.com/Speechproducts/speechproducts.htm

Lubert, J. & Campbell, S. (1998) Speech Recognition for Students with Severe Learning Disabilities From the Trillium Speech-to-text Project: A Manual for Implementing Dragon Dictate. http://snow.utoronto.ca/best/special/lessons/manual2.html

Miles, M., Martin, D. & Owen, J. (1998) A Pilot Study into the Effects of Using Voice Dictation Software with Secondary Dyslexic Students. Devon Education Authority (unpublished)

Nisbet, P.D. & Wilson, A. (2002a) Speech Recognition in Schools: Using ViaVoice. CALL Centre, University of Edinburgh, Edinburgh. ISBN 1 898042 22 5

Nisbet, P.D. & Wilson, A. (2002b) Speech Recognition in Schools: Using Naturally Speaking. CALL Centre, University of Edinburgh, Edinburgh. ISBN 1 898042 23 3.

O'Hare, E. & McTear, M. F. (1999)   Speech Recognition in the Secondary School Classroom: An Exploratory Study.   Computers & Education 33.1 (Aug. 1999): 27-45.

Sanderson A., & Smits, A. (2001) Improvements in Literacy: The Effect of Voice Recognition Software and Text-to-Speech on English and Dutch Dyslexic Students. 5 th British Dyslexia Association International Conference.

SQA (2005) Special Assessment Arrangements. SQA, personal correspondence.

home . about the conference . programme . registration . accommodation . contact