Speaking Frankly
Steer Clear of "Wrecking the Beach"
You say "speech recognition." The computer types "wreck the beach." Mistakes like this are common when dictating to your computer, but they also damage the credibility of speech technology in the users’ eyes (or ears). Boosting accuracy — reducing the rate of mistakes — is key to achieving the most productivity and the least frustration from your speech recognition software. The truism "garbage in, garbage out" applies here. The clearer your speech, the more accurate your software will be. If you mumble and slur your words, you’ll get errors all over. Even if your computer is more powerful than NASA’s. Besides the general caution of speaking clearly and precisely that we recommended in the last issue of Speech Technology, the following are some additional tips can be useful in helping obtain better (more accurate) results with any desktop speech software package.
- Retrain the computer. People’s voices change on different days and even at different times within the same day. You can’t make your voice sound the same all the time, but with some programs, you can re-enroll, making the computer retune its voice model to the way your voice sounds right now.
- Focus your eyes somewhere other than the screen. Dictating while looking at the screen can slow you down and cause confusion. Users who look at the screen keep feeling the need for the text to catch up with the dictation before they continue speaking. Most people get significantly better results if they focus their gaze elsewhere, or keep their eyes closed.
- Keep the microphone at a consistent position. Each time you use your computer, keep the microphone the same distance from your mouth. Also make sure that the microphone is not picking up your breathing. The best position for most headset microphones is between the corner and front of your mouth, about 1/2 to 1 inch away from your mouth. For most people, this is about the width of a thumb. Do a "thumb check" from time to time — put your thumb between the microphone and your mouth to confirm the microphone is in the proper position.
- Try a new microphone. All microphones distort the sound of your voice as it’s transmitted to the computer, but some microphones work better than others. Many people get significantly better accuracy by switching to a higher-quality microphone. Headset microphones, which are likely to move around more relative to your mouth, tend to give better accuracy than handheld microphones.
- Try a new sound card, or a USB microphone. Some sound cards introduce static or noise into your speech signal, making it harder for your software to understand what you say. This is especially common for sound cards that are built-in to the computer’s motherboard. A separate, stand-alone sound card can boost accuracy significantly. Also consider a USB microphone, which sends speech information through the USB port, bypassing the built-in sound card.
- Give the computer context. Speech software guesses what words you said from context as well as from the sound of your voice. If the words you say are similar to what you’ve said before, the program tends to guess what you say more accurately. Use the vocabulary learning features in your software program to teach the computer your writing style and what words you tend to use most often.
- Correct mistakes. Take the time to correct mistakes the computer makes, following the instructions in your speech software program. This will help the software improve as you use it. It is essential to making talking to your computer fast and easy.
- Reduce the number of background programs. When dictating into your speech software’s text window with no other programs running, your computer’s power is dedicated to processing your speech. If other programs are also open, or if you’re dictating into another program, your computer must divide its processing power between processing your speech and other tasks. Some programs that use processing power run in the background, without appearing in an on-screen window. Virus software and "reminder" alarm clock programs are among the programs in this category. Turning off background programs you aren’t using can improve both speed and accuracy.
Dan Newman is the author of Talk to Your Computer: Speech Recognition Made Easy and The Dragon NaturallySpeaking Guide (both from Waveside Publishing). He is president of the speech software firm Say I Can (http://www.SayICan.com) in Berkeley, California.