Speech Recognition for Less

If you're a regular reader of AnandTech -- and my articles in particular -- you may already know that I use speech recognition for writing the vast majority of my content. About five years ago, after a wonderful spree of typing like a madman on an article, my hands and fingers started going numb. Yes, I have the dreaded carpal tunnel syndrome. I tried to type less, and that helped a little, but what I really needed to do was cut out typing as much as possible. After reading about some options, I decided to try Dragon NaturallySpeaking. That was version 8, and I've never looked back.

A couple years later, version 9 was released and I began using that. I didn't notice any major improvements in accuracy or speed, but it did seem like it was a little better. I believe version 9 also added support for Office 2007's ribbon interface, but since I still have no problem using a mouse I haven't bothered with using Dragon to send commands much. Version 10 came out last year and I've been meaning to write a review for a while now. That keeps getting pushed aside, so rather than a full review I'm going to do a quick update on Dragon NaturallySpeaking 10.

Unlike version 9, version 10 definitely has some noticeable improvements. For one, it simply seems to be faster at recognizing speech -- Nuance said that the latest version is optimized for dual-core processors, which would certainly account for the difference. Accuracy is still a very nebulous concept, and I'm not sure if Dragon has gotten better with the latest version or if I just adapted to using Dragon over the years. Whatever the case, I'm quite happy with the level of accuracy NaturallySpeaking provides, and version 10 is the best so far.

The other major change with version 10 -- not initially, but as of March this year -- is that we finally have a version of NaturallySpeaking that works with 64-bit Windows. It was a long time in coming, and I'm not sure what the holdup was, but with the latest patch you can now use Dragon NaturallySpeaking 10 with Windows Vista 64-bit. Hooray for progress! Windows XP support is limited to 32-bit versions, however, and I haven't taken the time to test naturally speaking with Windows 7.

Some of you are probably wondering how Dragon NaturallySpeaking compares to the built-in speech recognition in Windows Vista. I discussed that in my speech recognition article several years ago, and frankly I still haven't been able to invest a significant amount of time in using Microsoft's "free" speech recognition. It looks like Microsoft may actually do a little better for sending certain commands to your computer, but in general the speech recognition interface they provide just isn't nearly as seamless as NaturallySpeaking. I haven't looked at Windows 7 to see if anything has changed, but I somehow doubt that I would actually make a switch at this point. It's a lot like using Windows and Microsoft Office; you can get open-source alternatives for free and run Linux and OpenOffice, but ultimately you find yourself wanting to go back to Microsoft.

So why this blog post now? I just received an email from TigerDirect with a link to Dragon NaturallySpeaking 10 Standard, which allows you to purchase the application for $50 instead of $100. 50% off a great application is certainly a worthwhile investment, and for those that are willing to jump through the hoops you get a $20 mail-in rebate. That brings the total cost down to $30. $30 for what is arguably the most important tool I use? Sign me up! (Incidentally, Nuance sent me the Professional version of Dragon NaturallySpeaking 10, but I pretty much limit myself to use of the Preferred feature set. You might want to check out the datasheet for specific details on what features may not work with the Standard version. Note that Microsoft Office is listed as a feature of Professional; you can still use Standard to dictate in Word, but you can't issue specific Word/Excel commands.) Note: The link above is a direct add to your cart, and TigerDirect says it's a "limited time offer", so I don't know how long the link will remain active. If you have an alternative deal for Dragon NaturallySpeaking, please post it in the comments!

The only thing you need beyond the software is a microphone, and my experience is that you don't even need a great microphone. However, there's more to it than just getting a microphone that works; if you plan to use speech recognition on a regular basis like me, you'll also want a microphone that's comfortable… and I prefer something that doesn't go over the top of my head, since I don't want to mess up my hairdo. ;-) I got the Sennheiser ME3 with Andrea USB Pod from emicrophones.com back in 2006, and it was definitely a nice upgrade to the Logitech headset I was using. You can find less expensive microphones with a similar style, but I would suggest getting some sort of USB sound pod and bypassing your soundcard; I don't know if it works that much better, but it's great if you ever have to use your microphone on a different PC. Since I routinely switch between my desktop and a laptop, the USB adapter is a godsend; integrated audio on some laptops can be particularly bad, and don't even think about using the built-in microphone on your LCD!

If you've never tried Dragon NaturallySpeaking, you really ought to. With a modern PC, it's really quite awesome being able to dictate to a computer and have everything come out accurately… well, at least with around 98% accuracy, give or take. My only real problem with NaturallySpeaking is that there are a few quirks I still encounter. It's nothing drastic, but sometimes the program crashes and any updates to my dictionary that I've created are lost and need to be reentered. I use a lot of custom phrases for computer products; for example the Gateway NV52 required that I put "envy fifty two = NV52" into my phrase list. I also have just about every AMD and Intel processor number in my list, so when I add 10 processors and forget to save my user files and Dragon crashes a few hours later, it can be annoying. My other gripe is with the "we need more hard drive space" dialog routinely appearing. Regardless of what I do, that dialogue seems to stick around and pop up every few weeks.

Perhaps the biggest challenge is that Dragon NaturallySpeaking is so accurate that you start to get lazy; homonyms can still throw the program for a loop, so if you don't read what it transcribes carefully, you're likely to come up with some grammatical/spelling errors from time to time. That's actually high praise for the program: it's so accurate that you come to expect everything to work out properly and stop paying attention. Just try not to do that when you're writing an important female, or the results might be a little embarrassing! (Ironically, Dragon just screwed up that sentence. I truly did say "email" and Dragon thought I said "female" -- see what I mean about embarrassing?)

Final tally: Over 1200 words and I only had to correct around ten actual mistakes (i.e. not counting mumbled words). Yes, Dragon NaturallySpeaking 10 truly can achieve 99% accuracy.