WholeWord Speech Recognition feature

Type of feature

WholeWord speech recognition (also known as Avaya Recognizer) is a licensed feature. You must have RTU licenses to use this feature.

About WholeWord speech recognition

WholeWord speech recognition enables the Avaya IR system to recognize and respond to spoken responses from callers. Speech recognition also recognizes touchtones. WholeWord speech recognition provides a whole word method to recognize caller speech by comparing an entire spoken word with a statistical model of the same word spoken by thousands of different people.

WholeWord speech recognition is well suited to tasks where the required input is digits (0 to 9 and "oh") or "yes" and "no" responses. Standard vocabularies for WholeWord speech recognition are available in the following languages:

• Australian English
• Brazilian Portuguese
• Canadian French
• Castilian Spanish
• Dutch
• French
• German
• Italian
• Japanese
• Latin-American Spanish
• UK English
• US English

WholeWord speech recognition has the following characteristics:

• Users can create and use custom grammars.
• All standard vocabularies for supported languages can recognize:
  • digits zero through nine
  • "oh" for zero where appropriate
  • "yes" and "no"
  • fixed-length connected-digit strings for a string length of one to ten digits (one to 24 digits for US English)
• Word spotting is supported.

  Word spotting is where the keyword does not have to be spoken in isolation.

• Phrase screening is supported.

  Phrase screening is the ability of the system to determine whether or not a candidate keyword is a close enough match to be declared a valid keyword.

• The WholeWord recognition engine supports both talk-off and barge-in interruption of a prompt.
• An NLSR proxy recognition type must be reserved for WholeWord recognition.
• The WholeWord recognition engine is capable of gracefully denying a request for recognition if the system resources are too low, even if there is an available channel.
• Dedicated NLSR server and port numbers are not required, and channel resources may be shared.

  In a prompt and collect node of an application running touchtone and WholeWord recognition in parallel, the application can accept either input and the system will not require separate port numbers or channels to be dedicated to each type of recognition.

• The WholeWord recognition engine can start up automatically after a start_vs command is executed.
• Multiple languages are supported on a single system simultaneously (performance permitting).

Software requirements

WholeWord speech recognition requires the following software:

• WholeWord Recognition Base package
• WholeWord speech recognition language packages

Setup considerations

Installation of the base package and language packages for WholeWord speech recognition is password protected and must be performed by Avaya or a certified third-party service provider. Avaya must also provision the RTU licenses for using this feature on the Avaya IR system.

Installation of each WholeWord language package includes the installation of the standard grammars for that language.

Using this feature

This feature is used by voice applications on the Avaya IR system. Avaya recommends using Avaya IVR Designer to develop applications with WholeWord speech recognition capabilities. For more information, see Avaya IVR Designer Help.

Note:
VoiceXML applications in Avaya IR cannot use WholeWord speech recognition. The VoiceXML feature currently only supports proxy speech recognition.

� 2003 Avaya Inc. All Rights Reserved.