Beginning January 1, 2013
Stop by the new site and take a look around.
Thursday, May 27, 2010
Thursday Spotlight: Ginger Hanson
Time has read hare. Whoops. That isn’t right! I said, “Tom has red hair.” Darn that speech recognition software, can’t it get anything right? Well, it can and usually does. But sometimes I mumble or the dog starts barking at the UPS truck. Any number of events can alter how well a speech recognition program works.
I’ve been using speech recognition software for about 10 years. I started with IBM’s ViaVoice program and segued into iListen when IBM left Mac users high and dry. Since I use a Mac, iListen was my only option. It wasn’t a good one and it had a lot of trouble with my Southern accent. ViaVoice had been easier to train and much more accurate. If I’d come to iListen first, I would never have stayed with speech recognition. Fortunately, iListen got into bed with Dragon NaturallySpeaking several years ago and created MacSpeech Dictate, providing Mac users with a fantastic voice recognition program.
The beauty of MacSpeech Dictate (and I presume, Dragon NaturallySpeaking) is how easily it learned my voice. These programs used to have a major learning curve, but now the user has a greater learning curve than the software. I’m still trying to master all the commands while MacSpeech Dictate merrily writes what I say with 95% accuracy.
Improving accuracy is important, but users new to speech recognition need to understand the program can’t do it alone. Five elements contribute to increased accuracy:
* A good microphone. The microphone plays a key role in the use of speech recognition programs because it has directional and noise canceling properties. For best results, use the microphone recommended for the type of operating system on your computer. Microphones are available in headset, desktop or wireless configuration.
* The position of the microphone also plays a role in accuracy. The headset model needs to be off to the side of your mouth so it doesn’t pick up your breathing which will be interpreted as speech. Proper placement of the desktop and wireless models also contributes to their accuracy. I use the wireless model with its included lanyard which puts the microphone in the same spot each time I use it. If I clip it to a shirt, the microphone’s location varies depending on the shirt’s neckline, which can affect accuracy.
* Level of noise in the room influences accuracy. During the setup process, the program measures the amount of background noise and adjusts itself accordingly. If you first set up your microphone in a quiet room, it will have trouble with your speech if you use it later with Meatloaf’s "Bat out of Hell" blaring. This problem is simple to correct, just rerun microphone setup to compensate for the new sound level.
* Computer volume also affects speech recognition. It’s best to always check the computer’s volume to ensure it’s not too high because a high volume can distort your voice which impairs accuracy.
* Clear and consistent speech improves accuracy. These programs are a continuous speech recognition product, which means they rely on the context of words within phrases to obtain accuracy. When using speech recognition, relax and speak in a normal conversational tone at a normal pace. Accuracy declines if you speak too quickly or slur your words. It’s also helpful to speak in phrases or sentences. This involves thinking before you speak which isn’t my strong suit, but it helps the program with context, which in turn improves accuracy.
There are two disadvantages to using speech recognition programs:
1. Speech recognition programs will probably always have problems with homonyms because this program is trying to guess what you’re saying. For example, “To be or not to be” came out accurately but “two bees in a jar” became “to be in a jar.” Since accuracy improves with context, the program has a better chance of figuring out which word to use when given the most context. And sometimes, it just won’t be able to figure the word out and you’ll have to type it in yourself.
2. Speech recognition relies on a consistent voice and just as a writer can wear out fingers and wrists by typing too much, a writer can wear out his or her voice by talking too much. To avoid vocal strain, keep drinking! No, not alcohol. Just something to keep vocal cords lubricated. Room temperature, plain water is best. Sorry to say, caffeine promotes dehydration, so avoid it while using a speech recognition program.
While these are powerful programs, they aren’t intended to completely replace the mouse and keyboard. Yet. As a long time Star Trek fan, I know that day is coming. For now, I’ve used speech recognition to write novels, essays (including this one), emails, etc. If you use the computer a lot and haven’t tried out speech recognition software, you might want to give it a whirl. I’m glad I stuck with it.