Integration with Google Speech for voice applications

In Avaya Experience Portal, applications like Orchestration Designer or VXML can request speech recognition from Google by defining a specific VXML grammar. The VXML grammar must have application/avaya-ep-csr type attribute.

Google Speech does not support standard grammar specifications such as SRGS or SISR.

The content of a VXML grammar for Google Speech is formatted as a CDATA section that encapsulates a JSON string. The JSON contains the properties that are related to this recognition request and are sent to Google through the Google Cloud Speech-to-Text API.

	Note:
	Only a single JSON string is supported within a grammar when a Speech-to-Text transcription is complete.

The following is an example of a VXML grammar for Google Speech:

<grammar mode="voice" type ="application/avaya-ep-csr" xml:lang=”en-us”>
			<![CDATA[
             { "provider": "google", "profanity-filter": false, "chunkSize": 8192, "phrases": [“work”,”avaya”] }
	    ]]>
</grammar>

Grammar property definitions for Google

“provider” – (Mandatory) The name of the cloud speech recognition provider. Currently there is only one valid value google. This is a required property for Experience Portal. It is not a Google property.

“chunkSize” – (Optional) The size in bytes of the audio segments that are sent to Google for recognition. In the example above, the 8kb value requests Experience Portal to send 8kb chunks of audio data to Google. This property can be given a default value when adding the Speech Server for Google. Note that changing the default value may influence speech recognition quality.

“profanity-filter” – (Optional) The filter that controls whether inappropriate content should be filtered out by Google. This property can be given a default value when adding the Speech Server for Google.

“phrases” – (Optional) A set of words or phrases that are likely to be spoken. Such a set of words and phrases can be used to customise Speech recognition to a specific context. This property can be given a default value per application. This is accomplished by adding a phrases parameter to the vendor parameters for Google in the application, in the following format:

phrases=[“work”,”avaya”]

Accessing speech recognition results

The speech recognition results for Google Speech recognition are available in VoiceXML through the shadow variables similar to other speech vendors.

The following is an example of the format of these result variables:

application.lastresult$.utterance