Search engine for phrase recognition based on prefix/body/suffix architecture

• US 5,832,428 A
• Filed: 10/04/1995
• Issued: 11/03/1998
• Est. Priority Date: 10/04/1995
• Status: Expired due to Term

• Current Assignee: Apple Inc.
• Original Assignee: Apple Computer Incorporated (Apple Inc.)
• Inventors: Hon, Hsiao-Wuen, Chow, Yen-Lu
• Primary Examiner(s): Hudspeth, David R.
• Assistant Examiner(s): EDOUARD, PATRICK NESTOR
• Application Number:
  US08/538,828
• Time in Patent Office: 1,126 Days
• Field of Search: 395/2, 395/2.63, 395/2.64, 395/2.65, 395/2.66, 395/2.6, 395/759, 395/760, 395/797, 395/798
• US Class Current: 704/254
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 15/1815 Semantic context, e.g. disambiguation of the recognition hypotheses based on word meaning
• 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

Abstract

A method of constructing a language model for a phrase-based search in a speech recognition system and an apparatus for constructing and/or searching through the language model. The method includes the step of separating a plurality of phrases into a plurality of words in a prefix word, body word, and suffix word structure. Each of the phrases has a body word and optionally a prefix word and a suffix word. The words are grouped into a plurality of prefix word classes, a plurality of body word classes, and a plurality of suffix word classes in accordance with a set of predetermined linguistic rules. Each of the respective prefix, body, and suffix word classes includes a number of prefix words of same category, a number of body words of same category, and a number of suffix words of same category, respectively. The prefix, body, and suffix word classes are then interconnected together according to the predetermined linguistic rules. A method of organizing a phrase search based on the above-described prefix/body/suffix language model is also described. The words in each of the prefix, body, and suffix classes are organized into a lexical tree structure. A phrase start lexical tree structure is then created for the words of all the prefix classes and the body classes having a word which can start one of the plurality of phrases while still maintaining connections of these prefix and body classes within the language model.

Claims

1. A method implemented in a digital processing system of constructing a language model in a speech recognition system, comprising:

receiving speech signals into a processor;

storing a plurality of phrases into a plurality of words in a prefix word, body word, and suffix word structure, wherein each of the phrases has a body word and optionally a prefix word and optionally a suffix word;

grouping the words into a plurality of prefix word classes, a plurality of body word classes, and a plurality of suffix word classes in accordance with a set of predetermined linguistic rules, wherein each of the respective prefix, body, and suffix word classes includes a number of prefix words of a first category, a number of body words of a second category, and a number of suffix words of a third category, respectively;

storing data elements representing interconnections among the prefix, body, and suffix word classes together according to the predetermined linguistic rules, wherein the language model generates signals representative of the received speech signals during a phrase-based search.

2. The method of claim 1, further comprising

identifying in the language model each of the body word classes a word of which can start one of the plurality of phrases;

identifying in the language model each of the body word classes a word of which can end one of the plurality of phrases.

3. The method of claim 2, further comprising

connecting a phrase starting node to all of the prefix word classes;

connecting the phrase starting node to all of the body word classes a word of which can start one of the phrases.

4. The method of claim 3, further comprising

connecting all of the suffix word classes to a phrase ending node;

connecting to the phrase ending node all of the body word classes a word of which can end one of the phrases.

5. The method of claim 1, wherein the predetermined linguistic rules are Chinese language rules.

6. The method of claim 1, further comprising the step of organizing the words in each of the prefix, body, and suffix word classes into a lexical tree structure and storing a representation of said lexical tree structure in a memory in said speech recognition system.

7. A method of organizing a phrase search in a speech recognition system having a language model comprising a plurality of prefix classes each having a number of prefix words of a first category, a plurality of body classes each having a number of body words of a second category, and a plurality of suffix classes each having a number of suffix words of a third category, wherein the classes are formed and interconnected together in accordance with a set of predetermined linguistic rules, wherein the method comprises the steps of:

receiving speech signals into a processor;

organizing a plurality of phrases into a prefix, body and suffix structure, wherein each of the phrases has a body word and optionally a prefix word and a suffix word;

organizing the words in each of the prefix, body, and suffix classes into a lexical tree structure;

creating a phrase start lexical tree structure for the words of all the prefix classes and the body classes having a word from which can start one of the plurality of phrases while still maintaining connections of these prefix and body classes within the language model, wherein the language model generates signals representative of the received speech signals.

8. The method of claim 7, wherein the step of creating a phrase start lexical tree structure further comprises

identifying in the language model each of the body classes a word of which can start one of the plurality of phrases;

grouping the words of all of the body classes that can start one of the plurality of phrases in the language model with the words of all of the prefix classes together to form the phrase start lexical tree structure while still maintaining the interconnections of these prefix and body classes set by the predetermined linguistic rules within the language model.

9. The method of claim 8, further comprising

connecting a phrase starting node to the phrase start lexical tree structure;

connecting the lexical tree structure of each of the suffix classes to a phrase ending node;

connecting ending of the body words in the phrase start lexical tree structure to the phrase ending node.

10. The method of claim 7, wherein the predetermined linguistic rules are Chinese language rules.

11. An apparatus for constructing a search engine for a phrase-based search in a speech recognition system, comprising:

means for receiving speech signals into a processor;

means for providing a language model for a plurality of phrases that are arranged in a prefix, body, and suffix structure, wherein each of the phrases has a body word and optionally a prefix word and a suffix word, wherein the language model includes a plurality of prefix classes each having a number of prefix words, a plurality of body classes each having a number of body words, and a plurality of suffix classes each having a number of suffix words, wherein the classes are formed and interconnected together according to a set of predetermined linguistic rules;

means for organizing the words in each of the prefix, body, and suffix classes into a lexical tree structure;

means for creating a phrase start lexical tree structure for the words of all the prefix classes and the body classes having a word which can start one of the phrases while maintaining interconnections of these prefix and body classes within the language model, wherein signals representative of the received speech signals are generated in response to the phrase-based search.

12. The apparatus of claim 11, wherein the means for creating further comprises

means for identifying in the language model each of the body classes a word of which can start one of the plurality of phrases;

means for grouping the words of all of the body classes that can start one of the plurality of phrases in the language model with the words of all of the prefix classes together to form the phrase start lexical tree structure while still maintaining the interconnections of these prefix and body classes set by the predetermined linguistic rules within the language model.

13. The apparatus of claim 12, wherein the means for creating further comprises

means for connecting a phrase starting node to the phrase start lexical tree structure;

means for connecting the lexical tree structure of each of the suffix classes to a phrase ending node;

means for connecting the phrase start lexical tree structure to the phrase ending node if the phrase start lexical tree structure includes at least one body class a word of which can end one of the phrases.

14. The apparatus of claim 11, wherein the predetermined linguistic rules are Chinese language rules.

15. A method of recognizing speech in a speech recognition system having a processor and a memory, said method comprising:

storing at least a portion of a language model in said memory, said language model for a plurality of phrases that are arranged in a prefix, body, and suffix structure in accordance with a set of predetermined linguistic rules, wherein each of the phrases has a body word and optionally a prefix word and optionally a suffix word, and wherein the language model includes a plurality of prefix classes each having a number of prefix words of a first category, a plurality of body classes each having a number of body words of a second category, and a plurality of suffix classes each having a number of suffix words of a third category, and wherein said classes are interconnected with interconnections in accordance with said set of predetermined linguistic rules;

receiving a speech utterance which includes at least one particular phrase;

performing an acoustic matching process on a representation of said speech utterance by using said language model.

16. A method as in claim 15 wherein said language model includes a data structure representing the words of all the prefix classes and the words of the body classes having words which can start said one particular phrase.

17. A method as in claim 16 wherein said data structure comprises a lexical tree structure which maintains said interconnections of said classes.

18. A method as in claim 15 wherein said step of performing an acoustic matching process comprises not matching certain words in said language model such that said acoustic matching process avoids matching said certain words, and wherein said processor performs said acoustic matching process.

19. A method as in claim 18 wherein said certain words comprise a plurality of certain suffix words.

20. A method as in claim 19 wherein said certain words further comprises a plurality of certain body words.

21. A speech recognition apparatus for recognizing speech, said apparatus comprising:

a transducer which receives a speech utterance which includes at least one phrase;

a processor coupled to said transducer to receive a digitized representation of said speech utterance;

a memory coupled to said processor, said memory storing at least a portion of a language model for a plurality of phrases that are arranged in a prefix, body, and suffix structure in accordance with a set of predetermined linguistic rules, wherein each of the phrases has a body word and optionally a prefix word and optionally a suffix word, and wherein the language model includes a plurality of prefix classes each having a number of prefix words of a first category, a plurality of body classes each having a number of body words of a second category, and a plurality of suffix classes each having a number of suffix words of a third category, and wherein said classes are interconnected with interconnections in accordance with said set of predetermined linguistic rules, and wherein said processor performs an acoustic matching process on said digitized representation of said speech utterance by using said language model.

22. An apparatus as in claim 21 wherein said language model includes a data structure representing the words of all the prefix classes and the words of the body classes having words which can start said one particular phrase.

23. An apparatus as in claim 22 wherein said data structure comprises a lexical tree structure which maintains said interconnections of said classes.

24. An apparatus as in claim 21 wherein said acoustic matching process comprises not matching certain words in said language model such that said acoustic matching process avoids matching said certain words.

25. An apparatus as in claim 24 further comprising

a display coupled to said processor, said display displaying text recognized by said speech recognition system and wherein said certain words comprise a plurality of certain suffix words.

26. An apparatus as in claim 25 wherein said certain words further comprise a plurality of certain body words.

27. A computer readable medium containing executable instructions which, when executed in a processing system, causes the processing system to perform a method for recognizing speech comprising:

receiving speech signals into a processor;

processing the received speech signals using a phrase-based language model wherein;

a plurality of phrases are stored into a plurality of words in a prefix word, body word, and suffix word structure, wherein each of the phrases has a body word and optionally a prefix word and optionally a suffix word;

the plurality of words are grouped into a plurality of prefix word classes, a plurality of body word classes, and a plurality of suffix word classes in accordance with a set of predetermined linguistic rules, wherein each of the respective prefix, body, and suffix word classes includes a number of prefix words of a first category, a number of body words of a second category, and a number of suffix words of a third category, respectively;

data elements representing interconnections among the prefix, body, and suffix word classes are stored together according to the predetermined linguistic rules; and

providing output signals representative of the received speech signals.

28. The computer readable medium of claim 27, wherein the method further comprises:

identifying in the language model each of the body word classes a word of which can start one of the plurality of phrases; and

identifying in the language model each of the body word classes a word of which can end one of the plurality of phrases.

29. The computer readable medium of claim 28, wherein the method further comprises:

connecting a phrase starting node to all of the prefix word classes; and

connecting the phrase starting node to all of the body word classes a word of which can start one of the phrases.

30. The computer readable medium of claim 29, wherein the method further comprises:

connecting all of the suffix word classes to a phrase ending node; and

connecting to the phrase ending node all of the body word classes a word of which can end one of the phrases.

31. The computer readable medium of claim 27, wherein the predetermined linguistic rules are Chinese language rules.

32. The computer readable medium of claim 27, wherein the method further comprises organizing the words in each of the prefix, body, and suffix word classes into a lexical tree structure and storing a representation of said lexical tree structure in a memory in said speech recognition system.