Browse
 
Tools
Rss Categories

DNN ASR Engine with Grammars

Reference Number: AA-02494 Views: 150 0 Rating/ Voters

Beginning with version 19.3.100, Lumenvox is including a new ASR engine technology, based on end-to-end DNN architecture.  The Lumenvox SRE service will support both our legacy ASR grammar engine and our new DNN engine.  By default, the new engine will be used for all recognitions, but via configuration, you can choose which engine to use on a case by case basis of desired.

The default behavior is controlled by the client_property.txt setting  GRAMMAR_ENGINE


A Meta tag within the grammar can be used to choose the ASR engine to use for each interaction.

Meta tag: GRAMMAR_ENGINE

Used to choose which ASR engine to use at the interaction level

  • 0 -  The value from the client_property.txt GRAMMAR_ENGINE will be used.
  • 1 - Force legacy engine
  • 2 - Force DNN ASR engine

  • Default
    : 0


    Example, the following grammar will force the use of the Legacy ASR engine.

    <?xml version='1.0'?>
    <grammar xml:lang="es" version="1.0" root="root" mode="voice"
             xmlns="http://www.w3.org/2001/06/grammar"
             tag-format="semantics/1.0"> 

    <meta name="GRAMMAR_ENGINE" content="1"/>

    <rule id="root" scope="public">
    <one-of>
      <item>red</item>
      <item>black</item>
      <item>grey</item>
    </one-of>
            </rule>

    </grammar>


    The DNN ASR engine uses an "end-to-end" architecture, meaning "phonemes" are not used in the traditional sense.  When using the DNN ASR engine, lexicons with phonetic spelling, or grammars within in-line phonetic spelling are ignored.  However, the DNN ASR engine is very good at recognizing foreign or unknown words, and generally does not need special handling to accommodate such words.
    A future version of the DNN ASR engine may accommodate custom phonetic spelling, so it is not recommended to remove lexicons or in-line phonetic spelling from grammars.  In cases where recognition is not doing well for grammars with extensive lexicons or in-line phonetics can be set to use the Legacy engine (as described above)