|
|
|
|
|
Overview
WholeWord speech recognition is available as an optional installable package on the Avaya IR system. WholeWord speech recognition allows you to write applications that prompt for and understand spoken input from callers. This feature supports a standard vocabulary that includes:
Processing involved during WholeWord speech recognition on the Avaya IR system includes a caller response, the action of the recognizer, and processing.
Caller response
During a Prompt and Collect action, the system prompts the caller for a response. If the Prompt and Collect action allows for voice input, the system locates and reserves a free recognizer resource.
The recognizer
The WholeWord speech recognition software is also called the recognizer. The recognizer compares the incoming speech sample to the word models indicated by the recognition type chosen in the Prompt and Collect action. The recognizer computes a score for each of the models. The score represents the likelihood that the incoming speech matches a word model indicated by the recognition type. The recognizer returns the most likely candidate to the specified variable. For IVR Designer applications, the variable is stored in the Save Response In field in the Prompt and Collect, Automenu, and Menu nodes. For more information, see the appropriate node description in the Avaya IVR Designer Help.
For the following example the variable is [CollectedDigits] for an IVR Designer application. If none of the candidates sufficiently match, the input is rejected and a "?" or an empty string value is returned to the script. For example, for a prompt of "1, 2, or 3" the application developer selects the US_1_3 recognition type, which is used for a response consisting of "1", "2", or "3." The caller's voice response to the prompt is compared to a model of the numbers "1", "2", and "3," and a score is computed for each number. Whichever score is highest determines the recognized result. A value of "1", "2, or "3" is returned in [CollectedDigits]. If no value scores high enough, the recognizer returns a "?"as the value. The application then decides what to do next, based on the value of the field. For strings of digits, the recognizer first determines up to four of the top scoring strings and their resulting scores. The recognition DIP then returns to the script the top scoring string whose length falls between minimum and maximum number of digits specified in the application. The string is then stored in [Collected Digits].
