Nuance Dragon NaturallySpeaking Premium 11.5 review

Anyone who spends any time using a PC will inevitably end up talking to it, so it’s only natural that software should exist to put such utterances to good use. This is the realm of speech recognition software and it’s one in which Nuance has been plugging away for many years. Its latest release for Windows is Dragon NaturallySpeaking 11.5 and although this edition is being sold as a boxed standalone product, it’s also a free update for users of last year’s version 11.

Despite what you may have seen in the movies, speech recognition has yet to reach the stage where computers can actually understand what someone is saying and act accordingly. Instead, such software relies upon matching parts of speech to information stored in a database and displaying the corresponding words on screen. The number-crunching required to transcribe what someone has just said in reasonable time is considerable, but the prodigious amounts of processing power offered by today’s PCs makes it not only possible, but also eminently practical.

You can import pre-recorded files for the software to transcribe itself

The leaps and bounds by which modern speech recognition software has progressed is most noticeable when setting up Dragon NaturallySpeaking 11.5. For best results, the software needs to be trained to recognise a particular individual’s speech patterns. In the early days of the application, this meant reading aloud into a microphone for the better part of an hour to achieve somewhat mediocre recognition accuracy. Dragon NaturallySpeaking 11.5, on the other hand, takes just five minutes to get a good ear for the way you speak and we dictated this review immediately afterwards with only a few mid-dictation edits.

Importing documents is a good way to bump up Dragon’s specialist vocabulary

Dragon NaturallySpeaking 11.5 can be trained further to improve its already impressive accuracy (and can also import documents to add specialist terminology to its vocabulary), but unless recognition is way off, simply correcting misrecognised words as they occur is an easier way to fine-tune performance. This is simply a matter of saying “correct” followed by the errant word or phrase and, if there’s more than one occurrence, each match is highlighted with a number that can then be chosen verbally. A numbered list of other words then pops up and you just choose the correct one by saying its number, saying the word again or spelling it letter-by-letter if all else fails. The system works well, as do the other commands for navigating around and editing text, although this latter option does require the use of a specific vocabulary. Thankfully, a context-sensitive Desktop sidebar shows a list of available commands for when you’re stuck for what to say, but most are obvious enough to remember after seeing them once.