|
|
|
|
|
Description
A recognition type allows you to indicate which keyword or words are possible selections for each prompt. All of the recognition types and words used by an application constitute that application's vocabulary. A complete recognition type includes:
The standard WholeWord speech recognition package provides several commonly used recognition types. In IVR Designer applications, the recognition type is selected through the Response tab of the Prompt and Collect, Announce, Menu, Automenu and Voice Capture nodes (for more information, see the appropriate node description in the Avaya IVR Designer Help).
Models and speaker independence
Each word of the system's vocabulary is represented by one or more mathematical models that contain the speech signal characteristics of the word. The speech recognition process compares a person's voice to the set of predeveloped speech models. Each model is constructed from thousands of samples of the spoken word. Regional accents and dialects associated with a particular language are incorporated into each model. Each model also includes a mix of male and female speakers so that the system recognizes callers of either gender speaking in any of those dialects with any of those accents. The recognition type determines which models the incoming speech is compared to.
Grammars
A grammar, which is selected by a recognition type, is a set of rules that specifies allowable vocabulary words and vocabulary word combinations at any one point in the script (for example, "four," "five," "six," "no"). While collecting spoken input, the speech recognition algorithm uses models and grammars to generate a list of candidates that most closely resemble this spoken input. The algorithm returns the most likely match to the script or, if no match fits, rejects the input. When input is rejected, the algorithm returns a "?" or an empty string to the script. This return message is the same for all of the languages supported. All grammars provided in the speech recognition package share the same set of models for words that are common to their package's vocabulary.
