Dynamic categories for a speech recognition system

• US 5,613,036 A
• Filed: 04/25/1995
• Issued: 03/18/1997
• Est. Priority Date: 12/31/1992
• Status: Expired due to Term

• Current Assignee: Apple Inc.
• Original Assignee: Apple Computer Incorporated (Apple Inc.)
• Inventors: Strong, Robert D.
• Primary Examiner(s): MacDonald, Allen R.
• Assistant Examiner(s): Sartori, Michael A.
• Application Number:
  US08/428,834
• Time in Patent Office: 693 Days
• Field of Search: 395/2, 395/2.1, 395/2.4, 395/2.52, 395/6, 395/2.55, 395/2.6, 395/2.66, 395/2.79, 395/2.84, 395/2.87, 381/41-45
• US Class Current: 704/243
• CPC Class Codes: G10L 15/193 Formal grammars, e.g. finite state automata, context free grammars or word networks
• CPC Class Codes: G10L 2015/228 of application context

Abstract

Maintaining dynamic categories for speech rules in a speech recognition system which has a plurality of speech rules each comprising a language model and action. Each speech rule indicates whether the language model includes a flag identifying whether the words in the language model is dynamic according to changing data in the speech recognition system. At periodic intervals, such as system initialization or application program launch time, for each flag in each speech rule which indicates that words in the language model are dynamic, the words of each of the language model(s) are updated depending upon the state of the system. Concurrent with the determination of acoustic features during speech recognition, a current language model can be created based upon the language models from these speech rules.

Claims

1. A method of maintaining dynamic categories for speech rules in a speech recognition system having an acoustic feature extractor which is separate from said speech rules, said method comprising the following steps:

a. creating a plurality of speech rules each comprising a language model and associated action to be performed in said speech recognition system, each said language model including a phrase list having a phrase comprising a sequence of words that may be recognized by a speech recognizer, each of said plurality of speech rules further comprising a dynamic flag identifying whether said words in said sequence of words are dynamic such that additional words may be added to the sequence of words and one of said words in said sequence of words is capable of being deleted from said sequence of words;

b. at periodic intervals during run time of said speech recognition system, for each dynamic flag in each said speech rule identifying that said words in said sequence of words are dynamic, dynamically determining said words in each said sequence of words based on data stored in said speech recognition system;

c. determining acoustic features in a sound sample, wherein said acoustic features are distinct from said speech rules; and

d. creating a current language model based upon each said language model from each of said plurality of said speech rules, and making said current language model available to the speech recognizer.

2. The method of claim 1 wherein said periodic intervals include when acoustic features in said sound sample are detected in said speech recognition system and prior to any recognition of words in said system.

3. The method of claim 1 wherein said periodic intervals include when application programs in said speech recognition system are launched.

4. The method of claim 1 wherein said periodic intervals include when relevant application program data are updated in said speech recognition system.

5. A method of generating a dynamic language model for a speech recognition system having an acoustic feature extractor, said method comprising the following steps:

a. creating a plurality of speech rules each comprising a language model, associated action to be performed in said speech recognition system and dynamic flag, said dynamic flag identifying whether words in said language model of said speech rule are dynamic such that additional words may be added to the language model and one word in the language model is capable of being deleted from said language model, wherein said plurality of speech rules are separate from said acoustic feature extractor;

b. at periodic intervals during run time of said speech recognition system, for each dynamic flag in each said speech rule identifying that said words in said language model of each said speech rule are dynamic, dynamically determining said words in said language model of each said speech rule based on data stored in said speech recognition system;

c. upon detection of speech, determining acoustic features in a sound sample, wherein said acoustic features are distinct from said speech rules; and

d. creating a current language model based upon each said language model from each of said plurality of said speech rules, and making said current language model available to recognizer in said speech recognition system.

6. An apparatus for generating a dynamic language model in a speech recognition system comprising:

a. means for creating a plurality of speech rules each comprising a language model, associated action to be performed in said recognition system and dynamic flag, said dynamic flag identifying whether words in said language model of said speech rule are dynamic such that additional words may be added to the language model and one word in the language model is capable of being deleted from said language model;

b. means for dynamically determining at periodic intervals during speech recognition system run time said words of said language model of each said speech rule comprising said dynamic flag identifying that said words in said language model of said speech rule are dynamic, said dynamically determining means retrieving data stored in said speech recognition system; and

c. means for determining acoustic features in a sound sample upon said detection of speech, wherein acoustic features are distinct from said plurality of speech rules;

d. means for creating a current language model based upon each said language model from each of said plurality of said speech rules; and

e. means for making said current language model available to a recognizer in said speech recognition system.

7. The apparatus of claim 6 wherein said language models in said speech rule comprise commands or categories, said speech rules further comprise a second flag identifying whether said language models in said speech rules comprise said commands or categories, and said means for dynamically determining said words in each said language model determines whether said second flag identifies that said words in said language model includes a category and if so, performing said retrieving said data from said speech recognition system.

8. An apparatus for generating a dynamic language model in a speech recognition system comprising:

a. a first circuit for creating a plurality of speech rules each comprising a language model, associated action to be performed in said recognition system and dynamic flag, each said dynamic flag identifying whether words in said language model of said speech rule are dynamic such that additional words may be added to the language model and one word in the language model is capable of being deleted from said language model;

b. a second circuit for dynamically determining at periodic intervals during speech recognition system run time said words of said language model of each said speech rule having said dynamic flag identifying that said words in said language model of said speech rule are dynamic, said second circuit retrieving data stored in said speech recognition system; and

c. a third circuit for determining acoustic features in a sound sample upon said detection of speech, wherein said acoustic features are distinct from said speech rules;

d. a fourth circuit for creating a current language model based upon each said language model from each of said plurality of said speech rules; and

e. a fifth circuit for making said current language model available to a recognizer in said speech recognition system upon a completion of said creating of said current language model by said fourth circuit.

9. A method in a speech recognition system having an acoustic feature extractor and which uses a current language model for recognizing speech upon the detection of utterances comprising the following steps:

a. creating a plurality of speech rules each comprising an associated language model, action to be performed upon the matching of words contained in said language model, and a flag identifying whether said words in said language model are dynamic such that additional words may be added to the language model and one word in the language model is capable of being deleted from said language model, wherein said plurality of speech rules are separate from said acoustic feature extractor;

b. at periodic intervals during run time of said speech recognition system, for each flag in each said speech rule which identifies that said words in said language model of said speech rule are dynamic, dynamically determining said words of each said language model of each said speech rule by retrieving current data stored in said speech recognition system at said periodic interval; and

c. creating said current language model upon determining acoustic features in a sound sample based upon each said language model from each of said plurality of said speech rules, and upon completion of said creating of said language model, making said current language model available to a recognizer for speech recognition in said speech recognition system in order to match said words contained in said current language model and perform an action contained in speech rules matching utterances contained in said sound sample, wherein said acoustic features are distinct from said plurality of speech rules.

10. The method of claim 9 wherein said periodic intervals include when acoustic features in said sound sample are detected in said speech recognition system.

11. The method of claim 9 wherein said periodic intervals include when application programs in said speech recognition system are launched.

12. The method of claim 9 wherein said periodic intervals include when relevant application program data are updated in said speech recognition system.

13. The method of claim 9 wherein said language models in said speech rule comprise commands or categories, said speech rules further comprise a second flag identifying whether said language models in said speech rules comprise said commands or categories, and said step of dynamically determining said words in each said language model includes the step of determining whether said second flag identifies that said words in said language model includes a category and if so, performing said retrieving of said data from said speech recognition system.