Merging of language models from two or more application programs for a speech recognition system

• US 5,651,096 A
• Filed: 03/14/1995
• Issued: 07/22/1997
• Est. Priority Date: 03/14/1995
• Status: Expired due to Term

• Current Assignee: Apple Computer Incorporated (Apple Inc.)
• Original Assignee: Apple Computer Incorporated (Apple Inc.)
• Inventors: Pallakoff, Matthew G., Reeves, Arthur Arlo, Rodarmer, Kurt W.
• Primary Examiner(s): MacDonald, Allen R.
• Assistant Examiner(s): Smits, Talivaldis Ivars
• Application Number:
  US08/403,594
• Time in Patent Office: 861 Days
• Field of Search: 395/2.81, 395/2.82, 395/2.84, 395/2.52, 395/2.6
• US Class Current: 704/275
• CPC Class Codes: G10L 15/18 using natural language modelling
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence
• CPC Class Codes: G10L 2015/223 Execution procedure of a spoken command
• CPC Class Codes: G10L 2015/228 of application context

Abstract

A speech recognition system operating on a computer system, which uses a single speech recognizer for all of the currently running application programs and provides a way of efficiently determining the proper destination application program for recognized speech. The speech recognizer uses a language model formed from the merging of the language models from two or more application programs. The merged language model includes data values indicating which application program'"'"'s language model was the source of the language model elements so that when those elements are recognized, recognition results can be directed to that application program.

Claims

1. A method of speech recognition for a speech recognition system operating on a computer system, comprising the steps of:

the speech recognition system generating a merged language model from first and second language models corresponding to first and second application programs, and including a data value for an element of the merged language model indicating that the element came from one of the first and second language models;

the speech recognition system using the merged language model to identify elements which match a speech signal received at the speech recognition system;

the speech recognition system, for each identified element, using the included data value to determine from which of the first and second language models the identified element came; and

the speech recognition system directing a recognition result to the application program corresponding to that language model.

2. A method of speech recognition for a speech recognition system operating on a computer system, comprising the steps of:

a first application program running on the computer system creating a first language model of elements to be recognized for the first application program;

a second application program running on the computer system creating a second language model of elements to be recognized for the second application program;

the speech recognition system merging the first language model and second language model into a merged language model, including for an element of the merged language model a data value indicating which of the first language model and second language model provided the element;

the speech recognition system receiving speech signals;

the speech recognition system searching the merged language model to identify elements which match the speech signals;

and for each identified element, using the data value to identify which of the first language model and second language model provided the element;

and directing a recognition result to one of the first application program and second application program corresponding to the identified one of the first language model and second language model which provided the element.

3. A speech recognition system operating on a computer system, comprising:

a merged language model created from first and second language models corresponding to first and second application programs, the merged language model including for an element of the merged language model a data value which indicates that the element came from one of the first and second language models;

a speech recognizer for receiving a speech signal and using the merged language model to identify elements of the merged language model which match the speech signal, and for each identified element using the data value to determine from which of the first and second language models the identified element came; and

the speech recognizer further adapted for generating a recognition result and directing the recognition result to the application program corresponding to the determined language model.