Speech Technology Magazine

DICTATION: How to Talk to Your Computer

Most of us have had limited experience conversing with our computers-besides cursing at them when they crash. To dictate most effectively with speech recognition software, you'll need to unlearn old habits and gain new skills.
By Dan Newman - Posted Aug 31, 1999
Page1 of 1
Bookmark and Share
Most of us have had limited experience conversing with our computers-besides cursing at them when they crash. To dictate most effectively with speech recognition software, you'll need to unlearn old habits and gain new skills. This article will give you some tips for using any dictation program. Speak Clearly
If you say "Jeet?" your friend will understand it as "Did you eat?" Computers, however, are less adaptable than people. To achieve accurate results from any dictation program, you must speak clearly. Pretend you're Dan Rather or Katie Couric reading the news, or imagine that you're giving a presentation to a small group. It may help to imagine that talking is giving the corners of your mouth a workout. Don't exaggerate your sounds or force them. Speak normally, just more clearly. Pronounce Each Word
When we listen to someone speak, our ears receive an unbroken stream of sound. We reconstruct words that are only partially heard-or left out altogether. Because our brains reconstruct missing sounds so easily (especially small, common words like "a," "the," and "of"), in everyday speech we tend not to pronounce every sound and word. It's just not necessary. For accurate transcription from your computer, it's important to make sure every word you say to the computer is pronounced, not "missing." A sentence that is perfectly understandable to a person might not be clear enough to a computer. This change in speaking style does not mean that you have to slow your natural pace. Just be sure to speak clearly and pronounce each word. Pause If You Like
Your friends might think it odd if during an animated conversation you halt in mid-sentence to gather your thoughts. When speaking to the computer, though, you can pause as long as you like, to think, take a break, or arrange your notes. The computer won't get bored waiting for you. Give It Context
Your accuracy will increase if you speak in complete sentences because the computer has more context to use in deciphering your sounds. Speaking one word at a time usually decreases accuracy. So concentrate, think of the right words, and speak them as a continuous stream. Leave Out the Fillers
Dictation software will usually mistake "uh" and "um" for "a," "of," and similar-sounding words. Unfortunately, since "uh" and "um" sound so much like these words, there is no way to make your software ignore them. To learn to stop saying "um," practice being silent instead of saying something to fill the space. When you feel an "um" coming on, just say nothing. Speak with Inflection
When talking to the computer, people tend to imitate the robotic voices of computers in old sci-fi movies. But using a flat monotone will reduce your accuracy as well as put you to sleep. Keep photos of friends next to the computer and pretend you're speaking to them. This will help you use more natural tones and will brighten your day, too. Breathe
Breathing fully and sitting straight will help you speak clearly. If you slouch while dictating, your lungs will be compressed and your voice constrained, making it harder for the computer to understand your words. Close Your Eyes
Dictating while looking at the screen tends to be distracting-it makes your speech more hesitant and less natural. Try dictating with your eyes closed. Most people get significantly better results this way. If your boss thinks you're taking a nap, try looking at the wall or at the photos of your friends. Be Alert
When it's 4:00 a.m. and you're still dictating the proposal that's due tomorrow, your recognition results will suffer as much as you're suffering. Tired people tend to mumble and speak without energy. You'll get your best results when you're most alert. Relax
Stress and tension change voice quality and degrade recognition accuracy. Frustration may kick in as you see mistakes proliferate. Frustration makes you more tense, changing your voice, which generates more mistakes, more frustration, and still more mistakes. Errors are inherent in computer transcription (as in human transcription). The computer takes its best guess at what you say, and it often guesses incorrectly. Correcting mistakes is part of the normal process of using speech recognition. When you learned to type, correcting errors became second nature. With the right attitude, correcting recognition errors can become just as routine. It helps to have a sense of humor about the computer's errors, and the computer is great at generating real howlers of mistakes. Speech recognition software tries to fit what you say into something that makes grammatical sense, if not literal sense. So its weird guesses often fit right into your sentence. You say "fresh squeezed lemonade" and the computer types it out: "fresh grease lemon aid." Don't try that beverage at home. Working at the computer is not especially good for your body. Humans were not made to sit in place for hours at a time, arms forward, fingers twitching rapidly. The repetitive motions of typing can cause discomfort and, in some cases, serious injury. It doesn't help that the intensity of computer work can lead people to neglect their bodies, posture, and physical positioning for hours at a time. Many computer users make an association, conscious or unconscious, between using the computer and being tense. Muscle tension can make your voice tense, changing its pitch and quality. Because dictation software tends not to respond as well to tense voices, achieving high accuracy takes some unlearning of the "computers = tension" equation. Working towards better results with speech recognition software can actually help you develop healthy habits. Good posture, rest, exercise, and meditation not only improve general well-being but also, amazingly, make NaturallySpeaking recognize your voice better. At last, computer use that encourages good health! Learning to Compose: Start by Reading
We all learned to type by copying printed passages. The best way to learn to dictate is to begin the same way-by copying. By dictating something already on paper, you can practice learning the voice software without having to worry about composing sentences at the same time. Choose a letter, memo, or e-mail message typical of the type of writing you do. Then read it into the computer as if you're speaking to another person. Say "comma," "period," and other punctuation as needed. As you read, pretend the computer isn't even there. Reading aloud will help you adapt to talking to the computer. After a bit of practice, add in a few sentences of your own thoughts. By starting to compose out loud interspersed with reading, you'll overcome any natural inhibitions you might have. Easy and Chatty
When learning to compose by voice, start with what's easy. Try dictating a few sentences about today's weather, make a list of things to do, or compose a chatty letter to a friend. Composing Complex Documents
Save more difficult documents for later, after you're comfortable composing simpler texts by voice. To dictate on complex topics, you'll need the skills of both speaking properly to the computer and composing intricate prose by dictation. It's best to practice these skills one at a time. Only after you're skilled at more casual, "chatty" compositions should you dictate documents that:
  • have difficult or complex subject matter
  • follow a detailed outline
  • use complex sentence structure
  • are aimed at a demanding audience, such as professional colleagues
  • will be widely circulated
Here are some further dictation tips that may be helpful when working on complex documents.
  • Speak in complete sentences, or at least complete phrases. Think of what you're going to say before you say it. Composing a sentence in your head before saying it will help you maintain your train of thought.
  • For complex writing, outlining is key. For a detailed letter, write or dictate a few words summarizing the main point and subpoints of each paragraph. When you're dictating, look at the outline. You'll naturally expand the outline to complete sentences as you speak.
  • When you're writing a first draft, capture your ideas as quickly as you can get them on the screen. Don't try to edit as you go-you can come back and edit later. Tell yourself to keep talking.
  • If you need extra motivation to get over dictation hesitancy, think of all you have to gain from writing out loud. Imagine boosting your output by two fold. Imagine filling the screen with text without having to type. Relax your hands, arms, and shoulders, and think how nice it is not to have to bang away at the keyboard.
  • As in any writing, keep your audience in mind. Imagine the particular person you're writing for to help the dictation flow. Composing aloud, your writing style may change. You might be pleased with your new spoken style, teach yourself how to compose aloud in your "old" style, or use a combination of dictating and revising to get the results you want. Use whatever method you prefer.
  • Using speech recognition can be a good way to get over writer's block. Imagine you're telling a friend what's next, then tell the computer.

Dan Newman, President of the speech software consulting firm Say I Can, is the author of Talk to Your Computer: Speech Recognition Made Easy and The Dragon NaturallySpeaking Guide. Both books are available online at SayICan.com, by phone at 1 (877) SAY-I-CAN, and in bookstores.
Page1 of 1