Windows Speech Recognition: Tips For Success

Randy Guthrie – Microsoft Academic Developer Evangelist
blogs.msdn.com/MIS_Laboratory

I have just finished a lecture tour in Kansas and Arizona and one of the more popular talks I have given is a demonstration of some of the new features and functionality in Windows 7. During the presentation I also throw in a few extras like Live Mesh, Microsoft Security Essentials, and Zune Smart DJ. One of the most awe-inspiring features in Windows 7 is the speech recognition application. When I demo it, more often than not I get applause at the end of the demo. Speech to Text has actually been around since Windows XP (as an add-in), but it was integral in Windows Vista and has been included in Windows 7. Its primary purpose is to enable computer users who either can’t or don’t know how to use a keyboard to be able to efficiently use a computer. To launch the speech application in Windows 7, click on the Start Button, then type the word “speech” in the search box. Speech Recognition will be the first choice in the programs area. I’ve got quite a bit of experience using Speech Recognition programs, and so I’d like to share some of my tips for making this feature work great.

Tip #1: Configure your mic / headset

When you first launch Windows Speech Recognition, it will start a wizard to enable you to configure your microphone. This is an important step as it will attempt to set the mic gain and levels. 

Tip #2 Take the tutorial

Out of the box Windows Speech Recognition is about 75% accurate, but it won’t understand commands very well. When you take the 20-minute tutorial, not only will you learn the various commands, but you computer is improving its ability to recognize your voice, which greatly improves its accuracy. If you skip this step, you can always go back and do it later. There are some additional readings you can do to improve Speech’s accuracy even greater. If you have a strong language accent, going through the tutorial and additional training scripts is essential to achieving high accuracy.

Tip #3 Double-check your microphone settings

Once you’ve configured your microphone, you won’t be prompted to reconfigure the next time you use Speech, but sometimes the settings get changed because of other things you do to your computer, so its always good to check, particularly if you are having lots of errors. One common problem for notebook computer users is that even when you have a headset plugged in, the built-in microphone remains set to the default. To check, right-mouse click on the speaker icon in the system tray image then click on “Recording Devices”. A list of microphones will be visible image; make sure your headset is in the list and it is enabled and set as the default communications device. You should also disable the built-in microphone if it has a green check by it as well. Also select your headset/microphone then right-mouse click and select “Properties”, then click on the “Levels” tab and make sure your headset recording level is set to at least 50%. This level is set when you do the speech configuration wizard, but can get changed. I tend to talk loudly, so I set it around 65% for me. 

Tip #4 Speak very clearly when dictating

The way I describe the best way to do speech recognition dictation is to exaggerate your pronunciation slightly and be a bit monotone. Slurring words is death to a speech recognition program. I also use a slightly different inflection when I give commands versus when I am dictating text. This seems to work better the longer I use the application because over time the speech engine improves its ability to recognize context.

Tip #5 Use Speech to correct translation errors

When you use the built-in correction tool, it updates the speech context library so it is much less likely to repeat that mistake again.  If you just select the incorrect word with your mouse and hit the delete key, Speech doesn’t have a chance to learn not to repeat the mistake.

Tip #6 Let Speech search your Documents folder

During the setup process, Speech will give you the option of having the files in the Documents folder scanned.  This processes teaches the Speech program about the contexts that you typically communicate in, increasing accuracy of the voice translation.  The more files in your documents folder, the more context it has to work with.

Tip #7 Work in a quiet environment

The more noise around you, the harder it is for the Speech program to hear accurately. Also if you are at work in a cubicle-type environment, talking to your computer can be annoying to those around you.  If I am in a large lecture hall using a lavaliere mic to amplify my voice, I usually turn it off to eliminate the voice echo from the hall speakers. 

Most of the Speech features can be accessed by right mouse clicking on microphone button in the speech control image.  Even for a person with great keyboard skills, accurate speech recognition can write a document between two to four times faster than typing.  This makes it great for brainstorming and getting ideas down on paper. Keep your eye on the screen however. Mistakes are much easier to spot and correct when they occur rather than after you’ve dictated several more paragraphs. 

Cheers!

-Randy