System and method for a language understanding conversational system

Abstract

A virtual assistant device recognizes multiple wake-up phrases. In response to a particular wake-up phrase the device sends speech audio to either a default or a third party virtual assistant server. A virtual assistant server can receive speech audio and an indication of which of multiple wake-up phrases was used and, accordingly, send the speech audio, or text recognized from the speech audio using automatic speech recognition, to a third party server. A response from the third party server can be voice audio or text for the virtual assistant server to synthesize distinctively corresponding to the wake-up phrase.

What is claimed is: 
     
         1 . A service for voice conversational interfaces comprising:
 receiving speech audio content indicating a wake-up phrase that identifies a knowledge domain;   performing speech recognition;   performing natural language interpretation, wherein the natural language interpretation uses the knowledge domain to determine a user intent;   taking a fulfilment action in response to the user intent; and   sending a response message in response to the fulfillment action.   
     
     
         2 . The service of  claim 1 , wherein the speech audio content is received through a runtime API. 
     
     
         3 . The service of  claim 1 , wherein the speech audio content comprises a user ID. 
     
     
         4 . The service of  claim 1 , wherein actions occur through repeated conversation turns in a session and the speech audio content comprises a session ID. 
     
     
         5 . The service of  claim 1 , wherein the fulfilment action is to conduct a dialog. 
     
     
         6 . The service of  claim 1 , wherein the fulfilment action is to request necessary slot data and add it to conversation state history. 
     
     
         7 . The service of  claim 1  wherein the fulfilment action provides lifelike conversational interactions. 
     
     
         8 . A language understanding bot framework comprising:
 a plurality of keyword recognition models that may be invoked simultaneously in a device;   a plurality of language understanding models, each representing a knowledge domain;   a configurable system of services enabled to receive a keyword recognition result that indicates which of the plurality of recognition models is activated and conditionally invokes a corresponding language understanding recognizer.   
     
     
         9 . The framework of  claim 8  further comprising runtime language generation that generates a response to a user according to the language understanding recognizer result, wherein the language understanding recognizer produces a result. 
     
     
         10 . The framework of  claim 8  further comprising a speech recognition service, which follows the keyword recognition result, that converts speech audio to text and provides the text to the language understanding recognizer. 
     
     
         11 . A non-transitory computer readable medium for storing code that is executed by the processor to cause a system to:
 receive speech audio content that includes a wake-up phrase   identify a knowledge domain based on the wake-up phrase;   perform speech recognition;   perform natural language interpretation, wherein the natural language interpretation uses the knowledge domain to determine a user's intent; and   take a fulfilment action in response to the user intent.   
     
     
         12 . The non-transitory computer readable medium of  claim 11 , wherein the processor executes code and causes the system to send a response message in response to the fulfillment action.