Method and apparatus for improving speech command recognition accuracy using event-based constraints

• US 6,345,254 B1
• Filed: 05/29/1999
• Issued: 02/05/2002
• Est. Priority Date: 05/29/1999
• Status: Expired due to Term

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: International Business Machines Corporation
• Inventors: Van Buskirk, Ronald E., Lewis, James R., Ortega, Kerry A., Wang, Huifang, Ballard, Barbara E., Nassiff, Amado
• Primary Examiner(s): Dorvil, Richemond
• Application Number:
  US09/321,918
• Time in Patent Office: 983 Days
• Field of Search: 704/270, 704/200, 704/272, 704/275, 704/231, 704/239, 704/240, 704/250, 704/255, 704/256, 704/257
• US Class Current: 704/275
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue
• CPC Class Codes: G10L 2015/228 of application context

Abstract

A method and system for improving the speech command recognition accuracy of a computer speech recognition system uses event-based constraints to recognize a spoken command. The constraints are system states and events, which include system activities, active applications, prior commands and an event queue. The method and system is performed by monitoring events and states of the computer system and receiving a processed command corresponding to the spoken command. The processed command is statistically analyzed in light of the system events and states as well as according to an acoustic model. The system then identifies a recognized command corresponding to the spoken command.

Claims

1. In a computer system for speech recognition operating at various states and running a program to perform various events, a method for recognizing a spoken command, comprising the steps of:

monitoring at least one of said events and states;

receiving a processed command corresponding to said spoken command;

analyzing said processed command according to at least one acoustic model to identify a probable acoustic match;

analyzing said processed command to identify a probable context match using a statistical model to analyze at least one of said events and states; and

providing a recognized command based on said probable acoustic and statistical context matches.

2. The method according to claim 1, wherein said events include at least one of a system control activity, an active application, prior command and an event queue.

3. The method according to claim 1, wherein past events and states are used to modify said statistical model according to a particular user'"'"'s command choice pattern.

4. In a computer system for speech recognition, a method for improving the accuracy by which a spoken command is recognized, comprising the steps of:

monitoring system events;

predicting a next event according to said system events using a statistical model to analyze said system events; and

estimating said spoken commands according to said next event.

5. The method according to claim 4, wherein said system events include at least one of a system control activity, an active application, a prior command and an event queue.

6. The method according to claim 4, wherein past system events are used to modify said statistical model according to a particular user'"'"'s command choice pattern.

7. The method according to claim 4, further comprising the step of receiving a processed command corresponding to said spoken command, wherein said predicting step further includes analyzing said processed command according to one or more acoustic models and said estimating step further includes identifying a probable command according to said one or more acoustic models.

8. A system for recognizing a spoken command, comprising:

means for monitoring at least one of system events and states;

means for receiving a processed command corresponding to said spoken command;

means for analyzing said processed command according to at least one acoustic model so as to identify a probable acoustic match and according to a statistical model acoustic for analyzing at least one of said events and states so as to identify a probate context match; and

means for providing a recognized command based on said probable acoustic and statistical context matches.

9. The system as recited in claim 8, wherein said events include at least one of a system control activity, an active application, a prior command and an event queue.

10. The system as recited in claim 8, wherein past events and states are used to modify said statistical model according to a particular user'"'"'s command choice pattern.

11. A machine readable storage, having stored thereon a computer program having a plurality of code sections executable by a machine for causing the machine to perform the steps of:

monitoring system events and states;

receiving a processed command corresponding to a spoken command;

analyzing said processed command according to at least one acoustic model to identify a probable acoustic match;

analyzing said processed command to identify a probable context match using a statistical model to analyze at least one of said events and states; and

providing a recognized command based on said probable acoustic and statistical context matches.