Accessibility & Inclusion
Software > Accessibility & Language

Voice Recognition Software

By Lasa Information Systems Team

There are a variety of reasons you might want to use voice recognition software. For example it can help people with restricted use of their hands to write, or enable people who can't touch type to produce documents more quickly. In this article an advice worker shares his experience of using Dragon Naturally speaking

"I work as an adviser at a centre specialising in housing and priority debt, and for the past few years I've been suffering from worsening pains in my hands and arms which have greatly reduced my ability to write and to a lesser extent to type. After considering various solutions to assist me at work we decided to try a computer system using voice recognition software.

I worked out a method that combined the voice system with a small amount of typing. During interviews and telephone conversations I use the keyboard to make brief notes then produce all my case notes and letters by voice. I use the program Naturally Speaking (version 3.0), developed by Dragon Systems, which employs continuous speech recognition, allowing me to dictate in phrases rather than one word at a time (the method used with earlier voice recognition software).

I've been working in this way for six months now and am very pleased with how well everything has gone. The accuracy of my system is very high (about 98%). However, though this level may seem impressive I feel that in practice it is only just good enough to make voice recognition a mainstream replacement for writing or typing, mainly because of the time needed to make corrections.

On the other hand, I can certainly dictate faster than I could type. If, like me, you are not a touch typist, there is the added advantage of looking at the screen rather than at the keyboard as you work.

In common with most users I use a microphone mounted on a headset; alternatives are lapel or desktop microphones. In addition, I use a separate headset connected to my telephone for phone calls - this leaves both hands free to type.

Fortunately, I work in an area off the main office with screens around my desk, and this keeps any disruption to other staff at a minimum.

Making corrections is straightforward but time-consuming, so it's essential for the system to be as accurate as possible. Improvements in accuracy come through training your system to recognise both your voice and your style of writing. Also, using a microphone and sound card that are particularly well suited to voice recognition also helps.

The prospect of having to use voice recognition as my main work tool was quite daunting so I decided to take the task of finding the most appropriate system seriously. However, I was disappointed by the lack of accurate and impartial information available for both software and hardware. Most of the specialist resellers I contacted seemed ill informed or unwilling to give specific information about software packages or components - they seemed concerned only with selling a "complete solution" or expensive training packages. Fortunately, I was able to get a lot of information from the Internet and this proved invaluable in helping me make an informed choice.

Apart from Naturally Speaking there are three other continuous speech recognition products including IBM's widely advertised Via Voice. The most in-depth review so far of these products was in the U.S. PC Magazine. In common with most tests it shows Naturally Speaking to be the most accurate product with only Lernout and Hauspie's Voice Express coming close.

The hardware requirements for using voice recognition software are pretty stiff. If your computer does not have enough power the words on the screen will start to lag behind your dictation. I have a 300 MHz Pentium II based computer with 128 Mb RAM which is generally considered to be the minimum specification for "real time" dictation. It may be worth investing in Intel's new Pentium III chip which has specific instructions for voice recognition - recent tests using Naturally Speaking show that these give a 25% speed improvement over a Pentium II at the same clock speed. It also seems likely that Pentium III will be the minimum requirement for making use of the next improvements in recognition accuracy.

For those wanting further information, the technical support section of Dragon's web site hosts an excellent discussion forum where users share information.

Update

IBM Via Voice and Dragon Naturally Speaking are now both owned by Nuance.  Lernout & Hauspie went out of business in 2001.