Specifying arbitrary words in rule-based grammars

• US 20020123876A1
• Filed: 12/30/2000
• Published: 09/05/2002
• Est. Priority Date: 12/30/2000
• Status: Active Grant

• Current Assignee: Intel Corporation
• Original Assignee: Intel Corporation
• Inventors: Pokhariyal, Shuvranshu, Gough, Corey, Davidson, Jason, Aundhe, Shirish, Hernandez, Thomas
• Granted Patent: US 7,302,381 B2
• Time in Patent Office: Days
• US Class Current: 704/1
• CPC Class Codes: G10L 15/193 Formal grammars, e.g. finite state automata, context free grammars or word networks

Abstract

A method and apparatus are described for specifying arbitrary words in a rule-based grammar, such as a context-free grammar. A wildcard identifier is used in a context-free grammar file in place of a predefined category of words. Artificial phoneme combinations that represent generic words in a speech engine'"'"'s vocabulary database, and that represent pronunciations of the predefined category of words are defined by rules. A speech engine uses the artificial phoneme combinations to determine acceptable words to listen to, and generates a results object comprising generic words corresponding to the artificial phoneme combinations, as well as non-generic words that do not exist in a given CFG, where the generic and non-generic words exist in the speech engine'"'"'s vocabulary database. The word having the highest confidence level that is not a generic word is selected as the word most likely to have been spoken by the user.

Claims

1. A method comprising:

creating a rule-based grammar having a wildcard identifier in place of a predefined category of words;

defining rules to produce artificial combinations of unique sounds in a language, where each artificial combination represents a pronunciation of the words in the predefined category, and represents a generic word that is defined in a speech engine'"'"'s vocabulary database;

generating a set of artificial combinations of unique sounds by substituting the wildcard identifier with the rules; and

in response to human speech specifying a wildcard word, determining a number of potential words spoken by the user by finding the generic words and non-generic words that phonetically match the wildcard word, and then assigning each of the words a confidence level.

2. The method of claim 1, wherein the rule-based grammar comprises a context-free grammar (CFG).

3. The method of claim 1, additionally comprising selecting a non-generic word having the highest confidence level.

4. The method of claim 1, wherein a unique sound in a language comprises a phoneme.

5. The method of claim 1, wherein said generating a set of artificial combinations of unique sounds by substituting the wildcard identifier with the rules comprises converting the wildcard rule-based grammar into a standard rule-based grammar.

6. A method comprising:

specifying a wildcard context-free grammar (CFG) which includes a wildcard identifier in place of a predefined category of words, each of which are defined in the speech engine'"'"'s vocabulary database;

specifying a set of rules that define artificial combinations of unique sounds in a language, where each artificial combination represents a pronunciation of the words in the predefined category, and corresponds to a generic word that is defined in a speech engine'"'"'s vocabulary database;

converting the wildcard CFG file into a recognized CFG grammar file by generating a set of artificial combinations of unique sounds based on the rules; and

in response to human speech having one or more spoken units, generating a results object having a number of generic words corresponding to artificial combinations appropriate to a given spoken unit, and having a number of non-generic words in the speech engine'"'"'s vocabulary database appropriate to a given spoken unit, each generic word and non-generic word having an associated confidence level.

7. The method of claim 6, additionally comprising querying the results object for a word having the highest confidence level in the speech engine'"'"'s vocabulary database.

8. The method of claim 6, wherein a unique sound in a language comprises a phoneme.

9. A machine-readable medium having stored thereon data representing sequences of instructions, the sequences of instructions which, when executed by a processor, cause the processor to perform the following:

create a rule-based grammar having a wildcard identifier in place of a predefined category of words;

define rules to produce artificial combinations of unique sounds in a language, where each artificial combination represents a pronunciation of the words in the predefined category, and represents a generic word that is defined in a speech engine'"'"'s vocabulary database;

generate a set of artificial combinations of unique sounds by substituting the wildcard identifier with the rules; and

in response to human speech specifying a wildcard word, determine a number of potential words spoken by the user by finding the generic words and non-generic words that phonetically match the wildcard word, and then assigning each of the words a confidence level.

10. The machine-readable medium of claim 9, wherein the rule-based grammar comprises a context-free grammar (CFG).

11. The machine-readable medium of claim 9, wherein a unique sound in a language comprises a phoneme.

13. The apparatus of claim 12, wherein the rule-based grammar comprises a context-free grammar (CFG).

14. The apparatus of claim 12, wherein a unique sound in a language comprises a phoneme.

16. The apparatus of claim 15, wherein the rule-based grammar comprises a context-free grammar (CFG).

17. The apparatus of claim 15, wherein a unique sound in a language comprises a phoneme.

19. The system of claim 18, wherein the unique sounds in a language comprise phonemes.

20. The system of claim 18, wherein the rule-based grammar is a context-free grammar (CFG).

21. The system of claim 18, wherein the speech engine comprises the conversion module.

12. An apparatus comprising:

at least one processor; and

a machine-readable medium having instructions encoded thereon, which when executed by the processor, are capable of directing the processor to;

create a rule-based grammar having a wildcard identifier in place of a predefined category of words;

define rules to produce artificial combinations of unique sounds in a language, where each artificial combination represents a pronunciation of the words in the predefined category, and represents a generic word that is defined in a speech engine'"'"'s vocabulary database;

generate a set of artificial combinations of unique sounds by substituting the wildcard identifier with the rules; and

in response to human speech specifying a wildcard word, determine a number of potential words spoken by the user by finding the generic words and non-generic words that phonetically match the wildcard word, and then assigning each of the words a confidence level.

15. An apparatus comprising:

means for creating a rule-based grammar having a wildcard identifier in place of a predefined category of words;

means for defining rules to produce artificial combinations of unique sounds in a language, where each artificial combination represents a pronunciation of the words in the predefined category, and represents a generic word that is defined in a speech engine'"'"'s vocabulary database;

means for generating a set of artificial combinations of unique sounds by substituting the wildcard identifier with the rules; and

in response to human speech specifying a wildcard word, means for determining a number of potential words spoken by the user by finding the generic words and non-generic words that phonetically match the wildcard word, and then assigning each of the words a confidence level.

18. A system comprising:

a conversion module to accept a wildcard rule-based grammar file as input, and to convert the wildcard rule-based grammar file to a set of artificial combinations of unique sounds in a language;

a speech engine to accept human speech having a wildcard word as input, and to determine a number of potential words matching the wildcard word, the potential words comprising a number of generic words corresponding to the artificial combinations of unique sounds in a language, and a number of non-generic words; and

a speech adapter to interact with the speech engine by querying the speech engine for potential words matching the wildcard word, and by returning the word most likely to match the wildcard word spoken by the user.