4.

4. Tricks for avoiding repeated errors

This page outlines a number of tricks which I use to avoid recognition mistakes. This page specifically outlines tricks for correcting mistakes where some word or phrase is repeatedly misrecognized. You should check How to Talk to NaturallySpeaking for general tips for getting good recognition.

Tip 1: Run the Vocbaulary Builder

If you have not already done so, make sure that you run the Vocabulary Builder. (See the topic What Does the Vocabulary Builder Do? for more information.) When run with appropriate text, the Vocabulary Builder will significantly improve your accuracy.

You should be the Vocabulary Builder samples of documents that you have written in the past which are similar to the types of documents that you will be writing in the future. For example, if you use Dragon NaturallySpeaking to write your email then you should try to dump into a file the contents of all of the email messages that you have sent recently. In addition, if you use Dragon NaturallySpeaking to write documents or specifications, then you should also include samples of previous documents or specifications that you have written when running the Vocabulary Builder.

If you are having trouble with repeated errors, then you can use the Vocabulary Builder to help. To do this, dictate or type some paragraphs which include samples of the words or phrases which you are having trouble recognizing. Then duplicate those paragraphs to or three times in a document and include that document next time you run the Vocabulary Builder.

(Remember that each time you run the Vocabulary Builder you need to include all the documents which you want considered when building a language model. Each time the Vocabulary Builder is run, it rebuilds your language model from scratch as described in the topic What Does the Vocabulary Builder Do?)

If you are having trouble with a particular word or phrase do not simply create a document which includes that word or phrase repeated over and over again. Including such a document when you run the Vocabulary Builder will not improve your recognition accuracy on that word. This is because the Vocabulary Builder considers the context of all of the words. For example, if you're having trouble with the word "hello" and you create a document which contains the word "hello" over and over again, then after running the Vocabulary Builder on that document you will get good accuracy only if you dictate something like "hello hello hello hello". But if you dictate "this is an example of the word hello", then your sample document will be of no help.

Instead, try to include lots of examples of the words which are having trouble with in context. This will be much more useful in improving the accuracy of Dragon NaturallySpeaking when dictating sentences and paragraph containing those words.

For example, I give lots of demonstrations of Dragon NaturallySpeaking and during those demonstrations I often use similar phrases. Every once awhile, after a demonstration, I correct my misrecognitions and save the text which I dictated into a file. Over many months, I have collected a file which contains lots of different phrases and paragraphs which I often dictate when giving demonstrations. And I use this text file as input when I run the Vocabulary Builder. When I give a demonstration, I never say exactly the same thing, but I do tend to use the same words and phrases so by using the Vocabulary Builder in this way I significantly improve my demonstration performance.

Tip 2: Train the word

When you run General Training after first training a user in Dragon NaturallySpeaking, this training process adapts the pronunciations of all the words in the vocabulary to match your speaking style. However, occasionally there are words in the vocabulary which you speak differently then expected. Often, training these individual words will help in their recognition.

For example, if I create a new user and and start dictating, the word "pause" will always appear as "cause" no matter how good my sound card. However, if I correct these misrecognitions, and click the Train button in the Correction Dialog to train the word "pause" then after two or three corrections, Dragon NaturallySpeaking stops mistaking the word "cause" for the word "pause".

Sometimes it is enough just use the Correction Dialog. But I recommend for repeated misrecognitions training the specific words which are giving you trouble. You can train specific words from either the Correction Dialog, or from the Vocabulary Editor by selecting words and then clicking the Train button.

A related issue is whether to train words individually or in phrases. I suggest that you do both. The main difference between training a word in a phrase is that you are training it as you speak it since sometimes when you speak a word in a phrase for sentence you do not completely pronounce all of the syllables. However, training a word individually will help if it is simply a matter of an incorrect pronunciation in the vocabulary compared the way you speak that word.

See When Does NatSpeak Adapt and Learn? for more information.

Tip 3: Add a phrase

Often, when I say the phrase "command and control", Dragon NaturallySpeaking types "commanded control". This is a very common type of miss recognition since I tend to eat the word "and" when I say the phrase "command and control".

A useful trick for solving common misrecognitions like this is to add the complete phrase to the vocabulary using the Vocabulary Editor. To do this, open the Vocabulary Editor from the Tools menu, and type in the phrase "command and control", and then click the Add button. Then, train this phrase. (Although it is not theoretically necessary to train words or phrases which you add to the vocabulary, I always recommend training them at least once says that will improve the pronunciation if the pronunciation the Dragon NaturallySpeaking gases is suboptimal.)

Dragon NaturallySpeaking tends to favor long phrases over short words, so by adding the phrase "command and control" to the vocabulary, it will be much more likely to be recognized even if you don't pronounce it completely accurately. This same technique can be used to force capitalization, for example I have added the phrase "Vocabulary Builder" with a capital "V" and capital "B" to my vocabulary so that whenever I dictate the phrase "Vocabulary Builder", it comes out capitalized correctly.

A number of people who regularly send me email have commented that they have added hundreds of phrases to their vocabulary which improves their accuracy significantly. Although, I do not go to this extreme, I am documenting the trick for people to try.

Tip 4: Use a different spoken form

Here is one final trick, the trick of last resort. If you just can't get Dragon NaturallySpeaking to recognize a word or phrase, then you can try using a dictation shorthand and say something different whenever you want a word or phrase typed. For example, let us say that you cannot get Dragon NaturallySpeaking to recognize "five and dime" even after including this phrase in the text which you feed into the Vocabulary Builder, and training this phrase, and even adding this phrase directly to the vocabulary. (All of those techniques should solve your problem.)

But for some reason either because the program is being troublesome or because you have a strong accent you cannot get Dragon NaturallySpeaking to recognize this phrase. What you can do is come up with a different phrase which you will say to get the troublesome phrase recognized. For example, you will associate the phrase "F A D" with "five and dime". Now using the Vocabulary Editor, add a new word whose written form is "five and dime" into spoken form is "F A D". Train this phase at least once.

Now, if you want to say something like " I went down to the five and dime and bought some new clothes", you can say that "I went down to the F A D and bought some new clothes". And Dragon NaturallySpeaking should translate F A D into five and dime. Clearly this is not as good as having Dragon NaturallySpeaking recognize your troublesome phrase in the first place, but for a common phrase which you are always having trouble with this can be a reasonable workaround.