SpeakQ Tips

Adapted from work done by R. Follansbee, Education Development Center, Newton, MA
Under NIDRR Grants from U.S. Dept. of Education

Listed below are some strategies for getting the best results when using SpeakQ. Some of these points may be relevant to other continuous speech recognition programs, but several are unique to the features of SpeakQ. You can find some of these points in the SpeakQ Help file, but you should review this page before you begin using the program or if you are supporting someone else who is learning to use the program.

Complete the speech recognition training

This is critical to get the best results with speech recognition. We cover this topic in another online document and in the SpeakQ Help file.

Choose the best speech recognition option for your use

Choose the best speech recognition option for your use. SpeakQ offers two options for dictation: (1) Speak and Select, and (2) Speak Continuously.

In the Speak and Select mode, the dictated word(s) do not appear directly in the document, but in the word choice box. You can look at the recognition choices offered and listen to them with text-to-speech before selecting one. Speak and Select offers a much more supported environment for dictating and is a good place to start in a couple of situations: if the user is unfamiliar with speech recognition, is younger, or has greater difficulty with literacy skills. Users in these categories often tend to dictate more slowly and in shorter language segments; the slower pacing of this mode and ability to review and confirm the recognized words before entering them into the document offers greater support in dictating.

In the Speak Continuously mode, the software operates more like standard speech recognition software—i.e., the user speaks and the recognized words appear directly in the document. This can be disconcerting for less experienced users or those who struggle with literacy. However, since Speak Continuously is integrated with word prediction and text-to-speech, even this mode still offers unique supports for reviewing text and correcting recognition errors.

You can easily try each of these dictation modes by choosing them in the Speech Recognition menu in WordQ to see which one works best for you. You can also find more information about these two modes in the SpeakQ Help file.

Do not speak as if having a conversation with the computer

Since you are speaking to write on the computer, this may lead you to try to talk to the computer as if you were speaking to another person. In conversation, the listener “fills in” any gaps in understanding when the speaker speaks indistinctly, mispronounces words, changes direction in mid-utterance, and so forth. Unfortunately, continuous speech recognition software is not yet as clever as having a live partner, and the software has to hear every word distinctly.

Enunciate words clearly

This strategy grows out of the point above and may be the most important one of all. Longer words such as “arithmetic” and “butterfly” are distinct enough that they will usually be recognized correctly. But, shorter words can cause problems. For example, if one dictates the word "and" indistinctly, the software may very well hear "an," “in,” or some other short word. There can also be a significant problem where the sound from the end of one word gets attached to the beginning of the following word, thereby confusing the software. For example, "if you" might be recognized as "a few" unless the words are enunciated clearly.

The need to enunciate clearly is all the more important with continuous speech because the sounds at the beginnings and endings of words affect those in surrounding words. Such natural speech characteristics can be confusing for continuous speech software. Therefore, one has to get in the habit of articulating words quite clearly—that is, in a manner that may seem unnatural for actual conversation. The Speak and Select mode of SpeakQ readily supports dictating in this way.

Dictate in shorter chunks

Use shorter chunks, such as clauses, phrases, even single words. This relates to the point above about enunciating more clearly. While you may be able to dictate an entire sentence without pausing, you might find that you get better recognition initially if you try to say fewer words at one time. The Speak and Select mode of SpeakQ is also particularly effective with this method. Saying single words can also be effective if the words are longer and more distinctive sounding. But, as we mentioned above, short, single words are easily confused by the software. For example, saying “a” as a single word will often trick the software into guessing similar sounding words (e.g., “hay,” “eight,” etc.). However, if you say short words as part of phrase or clause (e.g., “I saw a horse,” “with a new computer,” etc.), it will be easier for the software to recognize.

Think of dictating as “typing by voice”

If you think of typing as a means of text input, it may help you understand what is required to achieve better results with speech input. When you type, you have to follow certain rules. For example, you do not type incomplete words or insert meaningless words (like “um”). Also, if you hit the wrong keys while typing (e.g., instead of "tree," you mistakenly type the word "free"), the meaning of the sentence may be unclear or even ungrammatical, and the reader may be unable to interpret the meaning. However, mistakes happen in typing, and you have to correct these, just as you have to correct speech recognition errors.

[ Page Top ]

Think first before dictating

Before you actually dictate to the computer, compose the next sentence in your head, or possibly even aloud. One should only turn on the microphone when ready to dictate

Do not begin speaking until you know how to end.

[ Page Top ]

Set a dictation pace that allows you to attend to the above

You should speak somewhat more slowly than would be expected in conversation. You should keep in mind that the model for dictation is different from normal conversation, and think about speaking phrases or clauses rather than entire sentences, in order to maintain an optimal pace of dictation. It may take many hours of practice during actual dictation you to become completely comfortable with this style of dictation. Again, the Speak and Select mode of SpeakQ readily supports dictating in this way.

[ Page Top ]

Take frequent breaks when dictating

You should never dictate for more than 30 minutes at a time (less when starting out) without taking a break for a drink of water, to get up and stretch, etc. Even though you may be used to speaking for longer periods, dictation is in many ways more demanding than casual conversation. Continuous speech is a cognitively demanding activity that can leave one tired, tense, etc., and frequent breaks, even if brief, are critical. When using SpeakQ, you can take breaks from dictation by writing using only word prediction and text-to-speech.

[ Page Top ]

Preview and train any new content words or proper names to be used

Even though many of the words you need may be in the dictionary, you will find it helpful to “prime” the system with new vocabulary by seeing it written first and saying it aloud during training. This sort of preview can take place as part of a pre-writing activity, which is also helpful for understanding and using the concepts, etc. Preview words by simply dictating them in a list, or through a separate assignment such as a vocabulary activity such as writing definitions for key concepts, etc. In SpeakQ, the words you want to preview can be entered into a training file for dictating before writing (covered in a separate document and a SpeakQ help file).

[ Page Top ]