Formatting Voice Commands

There are some tricks you can use to format voice commands. Consider the following problem:

You would like to give a command that changes your television to a specific channel.

You could do this by creating a voice command for every possible channel, but that's impractical. Instead, your voice command can contain ranges of words. The recognition engine will accept any one of the words. To create a command that changes your television to any channel would look like this:

TV channel (0|1|2|3|4|5|6|7|8|9)+

This will accept any command like "tv channel 0 1", or "tv channel 1 3 6". You can substitute the actual word for the number in the command like:

TV channel (zero|one|two) etc.

It works the same either way.

Note that nested “()” and “[]” are not allowed. The following will not work: (hello|(bill|sue))

See the voice command section in chapter 6 for information on how to control IR devices with voice commands.

The string expression you supply can include square bracket characters ([ ]) to indicate optional words and vertical bar characters, (|) to indicate alternative strings. Alternates must be enclosed in parentheses. For example, "(hello [there] | hi)" tells the speech engine to accept "hello," "hello there," or "hi" for the command. Remember to include appropriate spaces between the text that is in brackets or parentheses and the text that's not in brackets or parentheses. You can use the star (*) operator to specify zero or more instances of the words included in the group or the plus (+) operator to specify one or more instances. For example, the following results in a grammar that supports "try this", "please try this", "please please try this", with unlimited iterations of "please":

"please* try this"

The following grammar format excludes "try this" because the + operator defines at least one instance of "please":

"please+ try this"

The repetition operators follow normal rules of precedence and apply to the immediately preceding text item. For example, the following grammar results in "New York" and "New York York", but not "New York New York":

"New York+"

Therefore, you typically want to use these operators with the grouping characters. For example, the following grammar includes both "New York" and "New York New York":

"(New York)+"

Repetition operators are useful when you want to compose a grammar that includes a repeated sequence such as a phone number or specification of a list of items:

Although the operators can also be used with the optional square brackets grouping character, doing so may reduce the efficiency of Agents processing of the grammar.

You can also use an ellipsis (…) to support word spotting, that is, telling the speech recognition engine to ignore words spoken in this position in the phrase (sometimes called garbage words). When you use ellipses, the speech engine recognizes only specific words in the string regardless of when spoken with adjacent words or phrases. For example, if you set this property to "[…] check mail […]", the speech recognition engine will match phrases like "please check mail" or "check mail please" to this command. Ellipses can be used anywhere within a string. However, be careful using this technique as voice settings with ellipses may increase the potential of unwanted matches.

When defining the word grammar for your command, include at least one word that is required; that is, avoid supplying only optional words. In addition, make sure that the word includes only pronounceable words and letters. For numbers, it is better to spell out the word than use an ambiguous representation. For example, "345" is not a good grammar form. Similarly, instead of "IEEE", use "I triple E". Also, omit any punctuation or symbols. For example, instead of "the #1 $10 pizza!” use "the number one ten dollar pizza". Including non-pronounceable characters or symbols for one command may cause the speech engine to fail to compile the grammar for all your commands. Finally, make your voice parameter as distinct as reasonably possible from other voice commands you define. The greater the similarity between the voice grammars for commands, the more likely the speech engine will make a recognition error. You can also use the confidence scores to better distinguish between two commands that may have similar or similar-sounding voice grammar.