Meaning token dictionary for automatic speech recognition

• US 6,963,832 B2
• Filed: 10/09/2001
• Issued: 11/08/2005
• Est. Priority Date: 10/09/2001
• Status: Expired due to Term

• Current Assignee: Hewlett Packard Enterprise Development LP (Hewlett-Packard Enterprise Company), Valtrus Innovations Limited (f/k/a Dolya Holdco 9 Limited) (Key Patent Innovations Limited)
• Original Assignee: Hewlett-Packard Development Company, L.P. (HP Inc.)
• Inventors: Vanhilst, Michael
• Primary Examiner(s): ABEBE, DANIEL DEMELASH
• Application Number:
  US09/974,645
• Publication Number:
  US 20030069730A1
• Time in Patent Office: 1,491 Days
• Field of Search: 704/9, 704/255, 704/275
• US Class Current: 704/9
• CPC Class Codes: G10L 15/1815 Semantic context, e.g. disambiguation of the recognition hypotheses based on word meaning

Abstract

Systems and methods of automatic speech recognition are described. In one aspect, an automatic speech recognition system includes a speech recognition dictionary and a speech recognizer. The speech recognition dictionary includes a plurality of meaning tokens each associated with one or more pronunciations of one or more vocabulary words and signifying a single meaning. The speech recognizer is configured to convert spoken input into a sequence of meaning tokens contained in the speech recognition dictionary and corresponding to a sequence of vocabulary words most likely to have been spoken by a user.

Claims

1. An automatic speech recognition system, comprising:

a speech recognition dictionary comprising a plurality of meaning tokens, wherein a single meaning token has a same meaning associated with plural different spoken words that have different pronunciations but similar spoken meanings; and

a speech recognizer configured to convert spoken input into a sequence of meaning tokens contained in the speech recognition dictionary and corresponding to a sequence of vocabulary words most likely to have been spoken by a user, wherein different spoken inputs having different spoken words but similar meanings are converted into a same meaning token or same sequence of meaning tokens, and wherein at least one meaning token encodes one or more labels identifying one or more respective application-specific categories for use by an application program.

2. The system of claim 1, wherein each meaning token is characterized by a unique spelling.

3. The system of claim 2, wherein the spelling of a meaning token facilitates extraction of meaning by a language analyzer.

4. The system of claim 3, wherein the spelling of a meaning token encodes the one or more labels identifying one or more respective application-specific categories.

5. The system of claim 4, wherein an application-specific category identified by a label encoded in the spelling of a meaning token is an object category, a place category, an event category, or an action category.

6. The system of claim 1, wherein multiple meaning tokens are associated with each of one or more polysemous vocabulary words contained in the speech recognition dictionary.

7. The system of claim 1, further comprising a language analyzer configured to extract meaning from the sequence of meaning tokens provided by the speech recognizer based upon a set of task-specific semantic rules.

8. The system of claim 7, wherein the language analyzer is a deterministic rule-based language analyzer.

9. The system of claim 7, further comprising an application command translator configured to select an action from a set of application-specific actions based upon the meaning extracted by the language analyzer, and to issue one or more commands to carry out the selected action.

10. The system of claim 1, wherein the speech recognition dictionary is a data structure stored in a computer-readable physical medium.

11. The system of claim 1, wherein the plural different spoken words include different phrases of words.

12. An automatic speech recognition method, comprising:

converting spoken input into a sequence of meaning tokens contained in a speech recognition dictionary and corresponding to a sequence of vocabulary words most likely to have been spoken by a user,

wherein the speech recognition dictionary comprises a plurality of meaning tokens, and a single meaning token has a same meaning associated with plural different spoken words that have different pronunciations but similar spoken meanings such that different spoken inputs having different spoken words but similar spoken meanings are converted into a same meaning token or same sequence of meaning tokens, and wherein at least one meaning token encodes one or more labels identifying one or more respective application-specific categories for use by an application program.

13. The method of claim 12, wherein each meaning token is characterized by a unique spelling.

14. The method of claim 13, wherein the spelling of a meaning token facilitates extraction of meaning by a language analyzer.

15. The method of claim 14, wherein the spelling of a meaning token encodes the one or more labels identifying one or more respective application-specific categories.

16. The method of claim 15, wherein an application-specific category identified by a label encoded in the spelling of a meaning token is an object category, a place category, an event category, or an action category.

17. The method of claim 12, wherein multiple meaning tokens are associated with each of one or more polysemous vocabulary words contained in the speech recognition dictionary.

18. The method of claim 12, further comprising extracting meaning from the sequence of meaning tokens based upon a set of task-specific semantic rules.

19. The method of claim 18, further comprising selecting an action from a set of application-specific actions based upon the extracted meaning.

20. The method of claim 19, further comprising issuing one or more commands to carry out the selected action.

21. The method of claim 12, wherein the plural different spoken words include different phrases of words.

22. A computer program for automatically recognizing speech, the computer program residing on a computer-readable medium and comprising computer-readable instructions for causing a computer to:

convert spoken input into a sequence of meaning tokens contained in a speech recognition dictionary and corresponding to a sequence of vocabulary words most likely to have been spoken by a user,

wherein the speech recognition dictionary resides on the computer-readable medium and comprises a plurality of meaning tokens, and a single meaning token has a same meaning associated with plural different spoken words that have different pronunciations but similar spoken meanings such that different spoken inputs having different spoken words but similar spoken meanings are converted into a same meaning token or same sequence of meaning tokens, and wherein at least one meaning token encodes one or more labels identifying one or more respective application-specific categories for use by an application program.

23. The computer program of claim 22, wherein the plural different spoken words include different phrases of words.