Method and system for limiting the number of words searched by a voice recognition system

• US 5,758,319 A
• Filed: 06/05/1996
• Issued: 05/26/1998
• Est. Priority Date: 06/05/1996
• Status: Expired due to Term

• Current Assignee: Qwest Communications International Incorporated (Lumen Technologies, Inc.)
• Original Assignee: MediaOne Group Inc. (Comcast Corporation)
• Inventors: Knittle, Curtis D.
• Primary Examiner(s): Hudspeth, David R.
• Assistant Examiner(s): Chawan, Vijay B.
• Application Number:
  US08/658,339
• Time in Patent Office: 720 Days
• Field of Search: 395/2.6, 395/2.61, 395/2.64, 395/2.52, 395/2.65, 395/2.54, 395/2.4, 395/2, 395/2.55
• US Class Current: 704/251
• CPC Class Codes: G10L 15/08 Speech classification or search
• CPC Class Codes: G10L 15/18 using natural language modelling
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models
• CPC Class Codes: G10L 15/19 Grammatical context, e.g. disambiguation of the recognition hypotheses based on word sequence rules
• CPC Class Codes: G10L 15/193 Formal grammars, e.g. finite state automata, context free grammars or word networks
• CPC Class Codes: G10L 15/197 Probabilistic grammars, e.g. word n-grams
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence
• CPC Class Codes: G10L 2015/085 Methods for reducing search complexity, pruning
• CPC Class Codes: G10L 2015/228 of application context

Abstract

A method and system for limiting the number of words searched by a voice recognition system having a predetermined language model containing a first plurality of words corresponding to all words recognizable by the voice recognition system in a predetermined application having at least one mode. The predetermined language model includes a list of all states that correspond to each of the first plurality of words. Based on the current mode of the application of the voice recognition system a subvocabulary of active words containing a second plurality of the first plurality of words is determined. A signal identifying the subvocabulary of active words is transferred to a recognizer which turns on only those states in the predetermined language model corresponding to the active words. Finally, a search is performed on a portion of the predetermined language model according to the subvocabulary of active words so that the computation time of the voice recognition system is reduced.

Claims

1. A method for limiting the number of words searched by a voice recognition system on a user speech input, the voice recognition system having a predetermined language model containing a first plurality of words corresponding to all words recognizable by the voice recognition system in a predetermined application having at least one mode, wherein the predetermined language model further including a list of all states corresponding to each of the first plurality of words contained in the predetermined language model, the method comprising:

determining a current active mode of the application of the voice recognition system;

determining a subvocabulary of active words containing a second plurality of the first plurality of words based on the current active mode of the application; and

performing a search on the subvocabulary of active words so as to reduce the computation time of the recognition of the user speech input.

2. The method as recited in claim 1 wherein determining the subvocabulary of active words further comprises generating an activation bit-vector having a plurality of bits corresponding to each of the first plurality of words contained in the predetermined language model, the plurality of bits for identifying the active words.

3. The method as recited in claim 2 wherein generating the activation bit-vector further comprises turning on the states corresponding to each of the second plurality of words that have been identified as active words.

4. The method as recited in claim 3 wherein the states corresponding to each of the active words include a "current" state and "next" state and wherein performing the search includes determining whether the "next" state of the "current" state is turned on.

5. The method as recited in claim 3 wherein the predetermined language model includes a desired percentage of the first plurality of words that will remain unpruned by the voice recognition system, and wherein determining the subvocabulary includes determining a target number of the active words for pruning based on the desired percentage and the number of states actually turned on.

6. The method as recited in claim 1 wherein determining the subvocabulary of active words includes determining an identification of the user, the user having predetermined active words assigned thereto.

7. A system for limiting the number of words searched by a voice recognition system on a user speech input, the voice recognition system having a predetermined language model containing a first plurality of words corresponding to all words recognizable by the voice recognition system in a predetermined application having at least one mode, wherein the predetermined language model further including a list of all states corresponding to each of the first plurality of words contained in the predetermined language model, the system comprising:

means for determining a current active mode of the application of the voice recognition system;

means for determining a subvocabulary of active words containing a second plurality of the first plurality of words based on the current active mode of the application; and

means for performing a search on the subvocabulary of active words so as to reduce the computation time of the recognition of the user speech input.

8. The system as recited in claim 7 wherein the means for determining the subvocabulary of active words further comprises means for generating an activation bit-vector having a plurality of bits corresponding to each of the first plurality of words contained in the predetermined language model, the plurality of bits for identifying the active words.

9. The system as recited in claim 8 wherein the means for generating the activation bit-vector further comprises means for turning on the states corresponding to each of the second plurality of words that have been identified as active words.

10. The system as recited in claim 9 wherein the states corresponding to each of the active words include a "current" state and "next" state and wherein the means for performing the search includes means for determining whether the "next" state of the "current" state is turned on.

11. The system as recited in claim 9 wherein the predetermined language model includes a desired percentage of the first plurality of words that will remain unpruned by the voice recognition system, and wherein the means for determining the subvocabulary includes means for determining a target number of the active words for pruning based on the desired percentage and the number of states actually turned on.

12. The system as recited in claim 7 wherein the means for determining the subvocabulary of active words includes means for determining an identification of the user, the user having predetermined active words assigned thereto.