Suppose that you are working on an application that callers use to report problems with telephone service. Part of your transaction involves asking callers for a telephone number. For a touchtone interaction, you could prompt "Enter your 10-digit telephone number." The caller can do one of three things: press 10 keys, press fewer than 10 keys, or do nothing.
For a similar speech recognition interaction, you could prompt, "Say your 10-digit telephone number." The caller can say anything at all in response (words or other sounds) or say nothing. The recognizer listens, then attempts to map the sounds to the appropriate grammar (US_DIG, with minimum and maximum digits set to 10), and come up with a series of digits. No matter what the caller says, the recognizer does the same thing: it listens to the sounds coming through the telephone, and attempts to map 10 digits, or using phrase screening, rejects the utterance.
What can be done to improve the application design in this example? What could increase the chance that callers will say what the recognizer is prepared to accept? Your application could first ask if the caller has the required information, by prompting: "Do you have your telephone number available?" Callers who say no can be transferred to the attendant, while callers who say yes can be prompted for the number.