Breaking the sound barrier

Voice technology systems – those which allow users to create text documents, enter numbers into spreadsheets and even conduct sessions on the Internet using the voice – are likely to greatly improve the productivity of lawyers. A year ago, I tested voice recognition products for The Lawyer, and concluded that "the rapid developments of these products in the last few months should make us all want to wait to see what the next releases will have to offer".

Since then, great strides have been made with discrete speech products – those in which you must pause between each word. But continuous speech products still have some way to go.

Discrete products available include Dragon Dictate, Kurzweil, and IBM Voice Type. Both Kurzweil and Dragon re-released their systems recently and last spring IBM released a Windows version of its extremely accurate Voice Type. To further enhance these products, separate companies have developed product add-ons such as Word Express, Kolvox, NCC and Shakespeare, which make major improvements. For example, Word Express, developed by All Voice, vastly develops IBM's Voice Type so that it can accept direct dictation and enables delegated voice correction. These alone increased productivity by more than 25 percent in my tests.

The only continuous speech product available is Philips. Even this is not yet truly available; the first phase will be launched at the Barbican exhibition this week and will allow digital dictation, the first stage to full speech recognition.

I have provided a comparison of the pros and cons of the packages. It is shown below.

But is this technology usable? The answer has to be a qualified yes. Prices have continued to fall and I suspect will continue to do so. Set-up and training times on all products is also shorter although the more time spent training the systems to your voice the better the recognition. All the systems I tested improved accuracy after training and became more accurate the more they were used.

Last year was pivotal for speech recognition. A lot of things came together, including the Pentium processor and 16MB of memory as a standard for an entry level machine.

Aside from the features that may appeal to personal preference, discrete speech recognition products mainly differ from each other in cost, time needed for training, speed at which the software responds and how much RAM is required to operate them.

Final selection may depend on the location of a reseller. It is important to know that the firm whose software seems most appropriate to you has a reseller in your area to help in the installation and train you in the use of the software in your offices.

Finally ask some searching questions. Are fee earners and staff ready? Is the technical infrastucture in place? Are fee earners willing to change working practices?

Training may take several weeks before any software will work at maximum efficiency. This may seem a drawback and does require commitment. But the pay-off is hands-free performance of key tasks, plus real cost savings in the reduced need for skilled support and faster completion of legal work.