Voice recognition is the holy grail of computing: the ability to walk around your room and dictate to your PC is an attractive one, however the real experience with voice recognition is far different. Processing words as they are spoken, particularly with the vast number of voices, inflections, and accents, is difficult for the human ear at the best of times, and computers have an even worse time of it, often throwing up gobbledygook or taking an inordinate amount of time to train.

The voice recognition industry has seen a considerable amount of consolidation, with Lernout and Hauspie buying Dragon Systems, which was in turn bought by ScanSoft, who also recently acquired Philips FreeSpeech. For consumers the choice has now narrowed to IBM ViaVoice and NaturallySpeaking, which is a pity because competition begets innovation.

On the plus side, voice recognition works better now than it ever has. I've been using voice recognition for seven years now, and have trialled various versions of IBM ViaVoice, Dragon NaturallySpeaking, Philips FreeSpeech and Lernout & Hauspie's consumer range, but version 10 of IBM's ViaVoice is the easiest and most functional piece of voice recognition software I've ever tested.

ViaVoice 10 has many new improvements including a new recognition engine, smoother integration into other applications, faster load times and it works much quicker, and more accurately than it ever has.

While not leaps and bounds ahead of Dragon NaturallySpeaking, ViaVoice 10 is still superior. The new recognition engine is great, but it also helps that PC hardware and components have improved markedly over the years. Still, trying to use a PC to recognise your voice requires a lot RAM and a very good processor, as it's a massively hungry operation. Even on a very good system you can find your RAM and processing power being drained. This doesn't necessarily make your system run slower, but it does somewhat kill your ability to multitask.

Although IBM ViaVoice 10 only comes with US and UK language packs I didn't find it difficult at all to dictate in an Aussie accent straight out of the box. In the past I've had many problems with speech recognition programs tailored for the British or American accent trying to recognise my Australian patois, but I was very pleased with the way ViaVoice was able to cope.

Using the software straight out of the box is not recommended particularly if you will be dictating for long periods of time.
It's a laborious task but we highly recommend completing all of the training lessons. These are short stories you read aloud which help the software build up your particular voice and speech pattern profile. As voice recognition is an ongoing process, the more you use the software the better it gets at recognising your particular mode of everyday speech.

ViaVoice can also analyse your documents to build up a more accurate profile of your speech patterns, and also to find words the software doesn't recognise. You can then train ViaVoice to recognise these words as they're spoken in the future.
Analysing your documents can take a while depending on how many documents you have, and it's where you'll notice the most marked drain on processing power. We recommend analysing no more than ten documents at time with ViaVoice 10.

We test voice recognition in three ways: straight out of the box; with minimal voice training; and then with maximum voice training. This is the first time that I've been really impressed by speech recognition at all three stages, to the point where I'm tempted to down my keyboard. The software isn't 100%, but when you can dictate 120 words per minute I don't mind correcting the odd mistake. It's a much faster way of writing.

ViaVoice 10 comes with SpeechPad, a word processor that's reminiscent of WordPad, but you can dictate into Office Word, or any other program for that matter.

For true hands-free computing you can use ViaVoice to navigate around your PC and applications, but I still feel it is best used in conjunction with mouse and keyboard.