Dragon Naturally Speaking is a voice recognition, and dictation program from ScanSoft. In essence, Dragon allows you to dictate documents, compose email, and run a variety of programs using your voice alone.

The latest version of Dragon is V6.0, and this release of the application represents a vast improvement over previous versions (my experience is with V6.0 Preferred - there are more cut down versions, as well as versions with more advanced features). Whereas older versions had a lot of trouble accurately transcribing text, V6.0 is a product that is usable with minimal training, and should give you a high degree of accuracy after very little initial training. Personally, after no more than ten minutes running through the initial training text, I am able to dictate text with about 90% accuracy.

Features

Dragon Naturally Speaking is far more than a voice activated typewriter. In fact, it is possible to run your entire computer using voice commands alone. Among the things you are able to do using this software:

  • Dictate documents into Word Processing applications - Microsoft Word, and Corel WordPerfect have integrated support for Dragon, allowing you to vocally carry out practically any function in these applications as easily as using a mouse or a keyboard.
  • Compose email, open and read email, and manage your Inbox. Dragon is able to read your email back to you, using Text-to-Speech technology.
  • Browse the internet, with the ability to follow links, and interact with web pages with no reduction in functionality, compared with mouse or keyboard users.
  • Dictate into a Dictaphone while away from your computer, then have Dragon transcribe this into text at a later time.

It could be argued that some of these features, for example navigating web pages, is more difficult using Dragon than by using a mouse - and this is probably true on many occasions. However when considering features such as these, imagine a quadriplegic, and the difficulty they currently face using a computer. Being able to browse the web, open applications, navigate around their desktop - without having to move at all - is a feature that turns what could have been a frustrating, and slow experience, into one that is more natural, and far more user friendly.

Using the Program

Dragon is a very simple program to use, and is quite intuitive. Once the application is installed, you will need to optomise your headsets volume, and run through a sound quality test. After these are completed, you will read through a short training text - no more than seven to eight minutes of speech. On completion of this, user files are created for your particular voice, and you're up and running.

Commands are simple and intuitive. If you need to open Microsoft Word, you simply say 'Open Microsoft Word'. Assuming the program has recognised the command correctly, Microsoft Word will launch. This will work for any application that has a shortcut on the desktop - as long as Dragon hears the 'Open' command, then is able to find the application name that you said, that application will launch.

You can access the Start Menu, navigate through it, and open applications this way as well.

Dictating into a word processor is fast - you can speak at up to 160 wpm - and straight forward. The main thing to remember, is that the application has no way of knowing whether a pause in speech is a break for a comma, a full stop, or simply a pause as you compose the next series of words. So you must also dictate all your punctuation, formatting and navigation. So if you were to speak:

(open quotes) how bright your garden looks (exclamation mark) (close quotes) said (cap) gandalf (full stop) (new line) (open quotes) yes (comma) (close quotes) said (cap) bilbo (full stop) (open quotes) i am very fond indeed of it (comma) and of all the dear old (cap) shire (semi colon) but i think i need a holiday (full stop) (close quotes)

your result would look like this:

"How bright your garden looks!" said Gandalf.
"Yes," said Bilbo. I am very fond indeed of it, and of all the dear old Shire; but I think I need a holiday."

Although that may look very complicated at first, it really does become natural quickly, and you will soon be dictating formatted text without the experience feeling unnatural. If you then decided that you (for some reason) wanted the document to be in a different font, you could simply say:

'Select all'
'Set font Times New Roman 12'

The entire document would be selected, then changed into the selected font, in the selected size. Easy!

The Accuracy Question

Since I've been using Dragon, in preparation for the rollout currently in process to many Judges in my workplace, I've been surprised to find how many people have had experience with a Dragon product at some time in the past. They've all said similar things about their experiences - "It's not very accurate." While that may have been the case in previous versions of Dragon (and I've not had the chance to test any version apart from V6.0), it doesn't have to be the case now. However, I believe part of the negative experience that many people have with this product, and others like it, is that they do not take the time to train it properly. Even with this new version, the first time you start it up and start speaking, you will come across errors. If you think about it for a second though, this is hardly surprising. Dragon has a vocabulary of around 250,000 words. Based on less than 10 minutes of hearing you speak, it is not going to be 100% accurate. You will see errors - common ones I see are:

  • Running words together - Dragon may think you have said "Dragon Naturally Speaking's a" when you spoke "Dragon Naturally Speaking is a"
  • Two words with the same sound, but different meanings - "wood", "would". Dragon will try to use context knowledge to get around these problems - however if both words could fit, it may choose the wrong one.
  • Uncommon words that sound similar to common words - "Gimli the Dwarf" becomes "Dimly the Dwarf"

And here is the most important point to note if you want to use an application such as Dragon to its full potential - you must train it as it makes mistakes.

If Dragon makes a mistake, you are able to tell Dragon to correct the word or words it has transcribed wrongly. If you do this - choosing the incorrect words, and replacing them with the correct words through Dragon's Spell feature - Dragon will update your user files, and modify itself to your particular way of speaking. The next time you speak those words, it's highly likely that Dragon will transcribe correctly. The more you do this as you dictate, the more accurate Dragon becomes over time. Basically, the more you correct, the less you will need to correct. If you do not do this - if you simply delete the incorrect words, and replace them with the correct one, using Dragon will become a more and more frustrating experience for you. Correcting mistakes properly is not an option - it's vital.

The other factor that will contribute to your accuracy is the way you speak. You don't have to speak to Dragon like you're speaking to a child - but on the other hand, you can't expect it to be accurate if you speak like you would in normal conversation. If you speak clearly, try not to mash your words together in your sentences, and speak at a constant volume, Dragon will be perfectly happy. The Judges I have trained in Dragon are a great example of this - many of them are used to dictating into a Dictaphone, for their Associate to type up at a later point. When they spoke to Dragon in the same way, their first small paragraph spoken once they'd finished the initial training was 100% correct (which, incidentally, made it difficult to go onto train them in how to correct mistakes...)

The Proof is in the Pudding

Here is a short piece that I have dictated using Dragon - this is from a fresh install, about ten minutes training, and almost no correction following that training. What you read here is exactly as it was transcribed - I've corrected no errors. Formatting was applied as I spoke. Mistakes will be surrounded by a *, with what should have been transcribed noted at the bottom.

Prologue

1. Concerning Hobbits

This book is largely concerned with Hobbits, and from its pages a reader may discover much of their character and a little of their history. Further information will also be found in the selection from the Red Book of *West march*1 that *is*2 already been published, under the title of The *Hobbits*3. That story was derived from the earlier chapters of the *Reader Book*4, composed by *Bill bowed himself, the first top*5 to become famous in the world at large, and called by him There and Back *The gain*6.

Corrections:

1. Westmarch
2. Has
3. Hobbit
4. Red Book
5. Bilbo himself, the first Hobbit
6. Again

Following this, I have made corrections to the above mistakes, and re-read the main paragraph again. This is Dragon's second attempt - following correction, there is a single error. Every word that tripped Dragon up the first time around, was transcribed correctly the second time around. The single word it got wrong was transcribed correctly the first time, so this error may be able to be put down to the way I dictated it the second time.


This book is largely concerned with Hobbits, and from its pages a reader may discover much of their character and a little of their history. Further information will also be found in the selection from the Red Book of Westmarch that has already been published, under the title of The Hobbit. That story was derived from the earlier chapters of the Red Book, composed by Bilbo himself, the first Hobbit to become famous in the world at large, and called by him There and *Act* Again.





Sources:

Some technical details from www.scansoft.com/naturallyspeaking