ScanSoft Dragon NaturallySpeaking Preferred 8 review

Getting your PC to meaningfully accept voice commands or dictation has been a benchmark for futuristic computing since time immemorial. Dragon NaturallySpeaking (DNS for short) has been a leader in the field for some time now, and is rapidly becoming devoid of competition.

That doesn’t mean you can choose only DNS Preferred, though. At the bottom end, DNS Standard costs £44, while the DNS Professional Developer version, which includes extensive macro-scripting capabilities, is priced at £467. The latter features roaming user profiles, making your user files available anywhere on the network – a major plus for enterprise customers.

Compared to the Standard version, DNS Preferred has a few extra bells and whistles, such as support for cordless microphones, text-to-speech and dictation playback, plus digital recorder/PDA support. All three packages come bundled with a low-cost headset, bearing the Emkay brand. With a bendable mic boom, the headset is easily adjustable and produces good results, although the non-standard colour coding for the audio connections might throw you at first. Another complication before you even start in earnest is the new online activation process, which you’ll need to go through after five launches of the application.

Most of the improvements to DNS 8 lie under the hood, but the hood itself is beginning to look a little dated. DNS uses a garish dockable toolbar originating from last century and one that wouldn’t look out of place in Windows 3.1. No attempt has seemingly been made to integrate it with the look and feel of Windows XP, which is now a step that’s long overdue.

On the plus side, it’s good to see that a much wider list of mic/headset combinations are now supported, including array microphones and digital recorders. That also includes Pocket PCs and Palm PDAs. Another improvement is an increase in the number of English twangs it will recognise. As well as the usual US, UK and Australian, DNS 8 now adds Indian and South East Asian accented English to the list.

But the most significant improvement over the past few years is the speed of voice training. Whereas early speech-recognition tools took an hour or more of script-based training to get any kind of useful performance, we spent only five minutes with DNS 8 before we could achieve excellent levels of accuracy. Even particularly tough challenges, such as tongue-twisters, fared well. Application control has also been significantly improved in terms of flexibility, to the point that in a matter of minutes we were able to make Word do what we wanted without touching the mouse or keyboard.

To fully milk the potential of DNS 8, you need to spend more time training and some lessons can take 30 minutes or more to complete. You have to be quite rigorous when correcting mis-recognised words too, but after a day or so of moderate use you can easily attain recognition accuracy of more than 95 per cent. ScanSoft claims that this release is 20 per cent more accurate than before. What this actually means is that it now makes 20 per cent fewer errors, which in practice equates to only four words wrong in every hundred, instead of five. Nonetheless, this is a highly creditable and practical performance.

To further improve recognition scores, DNS 8 needs to know what you typically write about. It can comb through your email and documents to look for new words to further improve overall accuracy. If you have a lot of documents, this part of the process can take as long as 30 minutes. You can skip it but this isn’t recommended. All told, training took about 15 minutes on our PC, which is a decent trade-off for the added flexibility it gives. You can further fine-tune recognition by running the Acoustic Optimiser. If you have a large vocabulary, this can be a lengthy process, so it’s possible to schedule it for after-hours.