The early speech recognition tools required discrete speech. This meant that each individual word had to be spoken clearly on its own, with a space between each word. Continuous speech recognition was introduced a couple of years ago, and since then several companies have attempted to perfect it, with varying degrees of success. Lernout & Hauspie has upped the ante with this new product, Voice Xpress Professional version 4*.

With more than 260,000 recognised words, the company claims an accuracy rate of 98 per cent or more using the new software, with just a brief five minute learning period. 98 per cent is very high, especially for a continuous speech recognition product, but it does seem to be a realistic target. The initial draft of this review, which was dictated using the product, had an accuracy rate of around 90 per cent. It is easy to see how, with a little more training, that rate could be raised a further few per cent. This is particularly true since we only completed one five minute recognition test before beginning dictation, whereas Lernout and Hauspie recommends one or two more tests just to be sure.

In addition to dictation, this new package also allows users to control Windows applications, using phrases such as "Open Microsoft Word" and "Bold the next two words". This is quite an intuitive process, and although there is still a learning period, it can save time once you've got the hang of it. This feature is available in all versions of the product, but while the Professional version supports dictation and control in Microsoft's Word, Outlook, PowerPoint, Excel and Internet Explorer, the Standard version is purely for use in Word. This cut-down package also lacks the mobile recording tools.

You would expect that such a powerful piece of software would require a very powerful PC, but Voice Xpress Professional 4 does not need state-of-the-art hardware. Although performance is better on a Pentium III with 128MB of RAM, you can get away with far less. For example, part of this review was dictated on a 166MHz Pentium machine with 64MB of RAM, which is below the minimum specification. The initial 'enrolment' led to quite a bit of disk thrashing, but during dictation there were no such problems. And, since you can use the software with virtually any Windows word processor, to really save resources it is possible to dictate into Notepad and then cut and paste. Users with 233MHz Pentium or faster systems should have no problems at all.

*and yes, it did recognise its own name.

This is a powerful piece of software. Its recognition level is excellent, especially if you give it an hour of your time, yet it doesn't require a supercomputer. When you take into account features such as the program commands, and the ability to import dictation from recordings or WAV files, only two questions remain; is it possible to work in an office full of dictating people, and do people communicate their ideas as clearly through speech as they do when typing?