Browse Prior Art Database

Client Application Program Interface for Continuous Speech Recognition System

IP.com Disclosure Number: IPCOM000111926D
Original Publication Date: 1994-Apr-01
Included in the Prior Art Database: 2005-Mar-26
Document File: 2 page(s) / 63K

Publishing Venue

IBM

Related People

Booth, BJ: AUTHOR [+2]

Abstract

The following describes a client application program interface for the IBM Continuous Speech Recognition System (ICSRS).

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 52% of the total text.

Client Application Program Interface for Continuous Speech Recognition
System

      The following describes a client application program interface
for the IBM Continuous Speech Recognition System (ICSRS).

      An application program interacts with ICSRS in a client-server
fashion, where the application is the client and ICSRS is the server.

      A session with ICSRS is called a conversation.  This is
initiated with srStartConversation and concluded with
srEndConversation.  Within a conversation are a series of recognition
requests.  A recognition request starts with srListen which tells

ICSRS to begin listening to speech.  A recognition request is
completed with a call to srGetSpokenWords which retrieves the
recognized speech.

      ICSRS recognizes speech within a frame of reference called a
context.  Contexts are created in advance using the ICSRS
task-building utility.  The input to task building is a set of
sentences in Backus-Naur format that comprise the recognizable
utterances of the context.  Thus a context is a set of files produced
by the task-building utility.

      Contexts are made known to the conversation by defining them
(i.e., specifying that the context may be used in a recognition
request).  A context is defined using the srDefineContext function.
A number of contexts may be defined for a conversation.  From the
contexts defined to the conversation, only one is in use per
recognition request.  To use a defined context for a recognition
request, you activate it.  A context is activated using the
srActivateContext function.  One context must be active for each
recognition request, and it may be changed from request to request.
It is advantageous to use a series of small contexts (and switch
between them for recognition requests) because the recognition is
faster on smaller contexts.  A con...