DE Mechanism

The DE System follows a five-step process: (a) input of training data consisting of transcriptions and corresponding routing labels; (b) extraction of relevant keywords from training data for recognizer and NLU grammar lexicons; (c) generation of natural language grammars through combination of keywords with a unique garbage model; (d) training of NLU via derivation of semantic relationships between keywords; and (e) testing and tuning of grammars.

The recognizer grammar is loaded inside the 3rd party speech recognition engine and allows detection of specified keywords in a wide variety of natural language utterances. The NLU grammar is loaded inside the Vestec NLU engine and processes the output of the 3rd party recognizer for semantic interpretation and call-routing. The grammar development process is fully transparent and manipulable. Specifically, there is no use of conventional “black box” methodology that necessitates a laborious trial-and-error development approach.