Triggering the Dragon Voice recognizer

To preload a DLM at service scope, edit the service’s configuration file and then start the service.

Open the krypton.yaml file manually and specify the DLM in the preload section of the configuration.

Note: this must be done by editing the configuration file. You can view the preload configuration in Management Station but please do not edit these settings from Management Station because they will not be parsed correctly.

Note: Krypton modules (DLMs) are not cached across application restarts.

The preload section of the .yaml file uses the following format:

preload:
  - dataPack: 
      language:
      topic:
    objects:  
      - url: 
        weight: 
        type: 		

These fields are defined as follows:

Here's an example that preloads DLMs for three different domains in two languages, with one object defined for each language. Note that the values do not need to be surrounded by quotes or double-quotes but they are used here for legibility.

preload:
  - dataPack:
      language: 'eng-USA'
      topic:  'GEN'
    objects:  
      - url: 'http://host/path/finance-dlm.zip' 
        weight: 0
        type: application/x-nuance-domainlm	
      - url: 'http://host/path/airline-dlm.zip'
        weight: 0
        type: application/x-nuance-domainlm	
      - url: 'http://host/path/pizza-dlm.zip'
        weight: 0
        type: application/x-nuance-domainlm	
  - dataPack: 
      language: 'cmn-PRC'
      topic:  'GEN' 
    objects: 
      - url: 'http://host/path/finance-mandarin-dlm.zip' 
        weight: 0
        application/x-nuance-domainlm	 
      - url: 'http://host/path/airline-mandarin-dlm.zip'
        weight: 0
        type: application/x-nuance-domainlm	
      - url: 'http://host/path/pizza-mandarin-dlm.zip'
        weight: 0
        type: application/x-nuance-domainlm

For details about configuring the Krypton service, see Configuring Krypton.

To preload an NLE model at service scope, use Management Station to set the following properties in the Monitoring & Control tab, and then start (or restart) the service.

• resource.preloadOnStartup.id
• resource.preloadOnStartup.type
• resource.preloadOnStartup.uri

Note: resource.preloadOnStartup is an array of modules to load. Each item should be added as resource.preloadOnStartup[i] where [i] is the index number of the module you are adding.

Alternatively, you can add the properties to the User-nle01.properties file before starting. For details about configuring NLE, see Configuring NLE.

To load models for the duration of a session, load the contents of the manifest file by adding nlptype=config to the <grammar> element in your VoiceXML application.

<grammar src="http://base_path/?nlptype=config"/>

where

• base_path points to the central repository where you store the artifacts.
• nlptype is the trigger for using Dragon Voice recognition and interpretation.
• config is the trigger to load the base language model and semantic model (if configured in the manifest). You can load one semantic model during a session. It cannot be changed for the remainder of the session.

  For Krypton-only recognition, set nlps-audio-only or server.nlps.audioOnly (not shown here). You must not load a semantic model (and the manifest must not configure one).

  With additional key=value pairs, you can load built-in grammars and DLMs that are defined in the manifest. See Preloading builtins: session scope and Preloading DLMs: service scope.

Note: The <grammar> element points to the storage location but does not explicitly name the manifest file. The filename must be nuance_package.json. The NLP service automatically fetches the manifest and loads the engines.

To load a domain language model while loading the manifest, append the DLM name. You can load more than one DLM, and you can load additional DLMs later in the session.

Note: To avoid a duplicate objects error in Krypton recognition, don't load the same DLM under different names and/or URLs in one SIP session. Instead, refer to the DLM using the same URL and name or, more efficiently, refer to the same, already loaded DLM in multiple recognition requests within one SIP session.

This example loads the DLM with the default weight defined in the manifest:

<grammar src="base_path/?nlptype=config&1000_MainMenu"></grammar>

This example loads the DLM and overrides the weight defined in the manifest:

<grammar src="base_path/?nlptype=config&1000_MainMenu_weight=medium"/>

This example loads two DLMs by using ampersand (&) to concatenate their names: 

<grammar src="base_path/?nlptype=config&1000_MainMenu_weight=0.1&2000_BillingMenu_weight=0.25"></grammar>

For more on weights calculations, see Understanding weight.

Relevant excerpt from manifest file:

This example sets the weight of 1000_MainMenu and 2000_BillingMenu to 0.25 and 0, respectively:

   "krypton" : {
		"dpTopic" : "GEN",
		"dpVersion" : "3.7.1",
	
		"sessionObjects" : [
		 {
			"id" : "1000_MainMenu",
			"type" : "application/x-nuance-domainlm",
			"url" : "./A1000_MainMenu_DLM.zip",
			"weight" : 0.25		   
		 },
		 {
			"id" : "2000_BillingMenu",
			"type" : "application/x-nuance-domainlm",
			"url" : "./A2000_BillingMenu_DLM.zip",
			"weight" : 0		   
		 }
		],		
		…
   }

Note: Setting a weight of zero reduces the probability of words in the model to extremely low, but does not deactivate the DLM. To disable the model, do not load it.

To load a builtin while loading the manifest, append the name and assign a weight.

This example loads the DIGITS builtin associated with the base model and activates it with medium weight (0.25):

<grammar src="base_path/?nlptype=config&builtin_digits_weight=medium"></grammar>

This example loads and activates two builtins with explicit weights:

<grammar src="base_path/?nlptype=config&builtin_financial_weight=0.25&builtin_date_weight=0.5"></grammar>

For more on weights calculations, see Understanding weight.

Note: This discussion is about Dragon Voice builtins. There is no relationship to Nuance Recognizer built-in grammars. To see which builtins are available for your base language model, see the data pack Readme.