Dragon NaturallySpeaking vs Windows 7 Speech Recognition

Back in November, during National Novel Writing Month, I purchased Dragon NaturallySpeaking
to use for dictating my novels and other writing.  I got version 11 because Amazon.com had it for a great price. As a side note, I later picked up version 12 as a raffle prize for our TGIO party thanks to another sale. 

Anyway I have been pretty happy with it.  With its initial training, it works pretty well despite my Southern accent and tendency to talk to fast.  It works well enough that since I got it going on my desktop, I've used it to dictate at least 30-40% of my NaNo novel, all of my personal blog posts (including several lengthy ones with financial stuff and uncommon words), I dictated the last three posts on my Animania blog (yes, even with Japanese names!), and I used it for the “Is ColdFusion Dead?” post here back in November.

The desktop version offers a rich feature set, including various training dialogs you can read through to help it learn your speech patterns (seriously, go through the Dilbert one and try to keep a straight face!), the ability to scan your documents and emails to learn your typical writing patterns, and the ability to learn from the corrections you teach it. 

It isn't 100% perfect, of course, but correcting its mistakes is generally an easy process and it gets better as you become more comfortable using it and as you teach it.  Accuracy is also significantly enhanced by having a good quality headset, which is conveniently included with the software.  The included headset works great for dictation and is decently comfortable. It does tend to hurt my ears after an hour or so of straight use, but so does the phone so I don't hold that against it.

The only real issue I've had with the program is a tendency to get sluggish after an hour or so, however I also recognize this isn't entirely the program's fault.  I'm running Windows 7 64 bit with only 4 gigs of RAM – as you might expect, a speech recognition program can eat up some RAM.  When I upgrade to 8 gigs, I suspect the sluggishness issue will also but disappear.

One complaint often leveled about Dragon though is the cost.  The home edition is priced at $100, though Amazon often has it for around $50.00 and you can usually pick up the second most recent version for around $30.00.  Still that might be more than some people want a pay, especially when Windows 7 has a built-in free speech recognition program (it originally was included with Vista, but we like to forget Vista exists).

I'd seen a few folks comment about Windows built-in, home grown one and compliment it. A few even said it was better than Dragon.  So I decided to put it to the test on my work laptop. I doubt I'd ever get to the point of using it for coding, of course, but I could see it being useful for documentation purposes, so why not.  For the trial run, I used my Microphone headset, which I use for Lync calls.

Per the Speech Recognition area, I went through the tutorial.  I was already suspecting things were over hyped when the tutorial, which already knows exactly what you are supposed to say, kept asking “what was that” to my saying “Okay” or speaking specific sentences like it asked.  Still, I know my accent is a little wonky for computers for some reason, so I continued on.  The command set seemed impressive and the tutorial makes it seem fairly easy to use.  There were differences between it and Dragon, of course, but they weren't so drastic I couldn't adapt quickly.

When that was done, I went through a round of training. Like with Dragon, you read text from the screen in a “natural” voice so it can learn how you talk with known input.  However, even during the training, it frequently sat there while I said the displayed phrase or sentence two or three times before it finally understood me.

Once I was done there, I decided to put it to a real test – actual writing!  I started this post using the speech recognition.  However, I soon realized the system was way clunkier than Dragon.  In both programs, for example, you can correct something by saying “Select” followed by what you need to correct. In Dragon, it will immediately display a list of possible alternatives, based on how you said it at the time and the surrounding content.  90% of the time the correct words are already there, I just tell it which to choose and continue on.  This is particularly true when dealing with homonyms or just needing to fix casing. You can just say “Caps that” or “bold that” to do formatting.  And you can just speak a new bit of dictation to replace the selected text all together.

With Windows Speech Recognition, you have to speak to replace – no helpful “hey did you want this one” menu.  You can apply formatting right then, but if you want to fix bad mixed casing or if it just got the wrong “too”, you have the say it all over again.  Then it will offer you a list of options to pick from.  If what you want isn't on the menu, you can spell it out (like with Dragon) or say it again to get another menu of options.  Once you get the list, you say the number of the one you want, but Windows adds an extra step here as well – with dragon once you say “Choose X” it does it and moves on.  With Windows, you have to say “OK” after it highlights the selected number.

I found Windows spelling dialog to be clunkier as well. It could not get any capitalizations right unless I always put them in the form of “Capital A as in Apple” or “Capital D as in Dog”.  Capital A alone just didn't work.  This could, again, be it having issues with my accent, but considering Dragon gets it right the vast majority of the time, I'm thinking it's a weakness in the program.

The big deal breaker for me, though, that caused me to give up writing this entry in it after around the second paragraph, was how much it kept missing my saying even clear, basic works (like the whole missing OK during the tutorial).  And correcting some errors, like its writing Amazon.com as “amazon .Com” eventually had me turning to the keyboard. I could fix the casing, but getting rid of the space was apparently impossible. Other times when it added extra spaces, I could not select the space and delete it like I can with Dragon.

I'm sure Microsoft will continue to improve on the Speech Recognition system and hopefully for the better.  For now, though, I'd have to say the Windows 7 one is a pass for me.  With just trying to write fairly normal text, it was too clunky, too cumbersome, and too inaccurate.  I certainly can't imagine trying to use it for technical terms, character names, foreign words, or the like.