Using n-best

This section explains how to use Nuance n-best recognition processing in a VoiceXML application. With n-best processing, the recognition service generates an array of possible recognition results instead of just the single best solution. This allows you to operate on the results in various ways before returning an interpretation to the application. For example, you can:

• Add false recognitions to a skip list so as not to ask for confirmation of a mis-recognized word twice.
• Compare the results to some external application knowledge in order to determine which of the multiple results is valid.

  For example, when recognizing a credit card number, you can test if the recognized digit sequence is a valid account. If it isn’t, you can move on to a result that the recognizer considered to be less likely.

This technology has proven to increase accuracy when some form of application knowledge can be applied to a set of recognition results.