Homograph filter for speech synthesis system

• US 6,098,042 A
• Filed: 01/30/1998
• Issued: 08/01/2000
• Est. Priority Date: 01/30/1998
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Huynh, Duy Quoc
• Primary Examiner(s): Zele, Krista
• Assistant Examiner(s): Opsasnick, Michael N.
• Application Number:
  US09/016,545
• Time in Patent Office: 914 Days
• Field of Search: 704/260, 704/266, 704/4
• US Class Current: 704/260
• CPC Class Codes: G10L 13/08 Text analysis or generation of parameters for speech synthesis out of text, e.g. grapheme to phoneme translation, prosody generation or stress or intonation determination

Abstract

A homograph filter and method which increase the probability that homographs are pronounced correctly in a speech synthesis system utilizes a filter engine operating in conjunction with a set of rules. The filter engine parses a textual sentence to extract any present homographs and applies a correct set of rules to the homograph, based on an optimal search algorithm. The engine then carries out any appropriate substitution of phonetic data. Rules are primarily based on syntactic analisis, based on a priori knowledge of how each homograph is used. The rule set is classified into different categories in order to optimize the search algorithm and to allow the rules to be modified and updated incrementally without effecting the engine construction and/or performance. The search algorithm utilizes syntactic analysis to achieve optimum results. If syntactic analysis does not yield a satisfactory result, semantic analysis could also be utilized to determine the usage of the homograph based on the contents of the items which surround the homograph. The rule set contains a set of grammatical rules to perform syntactic analysis. If syntactic or semantic analysis does not yield a result, the result will be based on the statistical usage of the given homograph.

Claims

1. A computer program product for use with a computer system capable of converting text data into synthesized speech, the computer program product comprising a computer useable medium having program code embodied in the medium and configured to determine a preferred pronunciation of a homograph in the text data, the program code further comprising:

program code which examines the text data to identify the homograph within the text data and to extract words surrounding the identified homograph in the text data;

program code responsive to the identified homograph which identifies the possible parts of speech that the identified homograph can assume;

program code responsive to the possible parts of speech that the identified homograph can assume that obtains a set of rules, each rule based on a pair of possible parts of speech of the identified homograph and a word order and position of one of the surrounding words;

program code which sequentially applies the rules in the obtained rule set until a rule is satisfied to determine a part of speech for the homograph in the text data; and

program code which is responsive to the homograph and the determined part of speech usage for determining a preferred pronunciation for the identified homograph.

2. The computer program product of claim 1 wherein the program code configured to identify a homograph comprises:

program code configured to identify selected portions of the text data.

3. The computer program product of claim 2 wherein the program code configured to identify selected portions of the text data comprises:

program code configured to parse the text data; and

program code configured to delineate the text data into phrases.

4. The program code of claim 3 wherein the program code configured to delineate further comprises:

program code configured to identify punctuation characters peculiar to the natural language of the text data.

5. The computer program product of claim 2 wherein the program code configured to identify a homograph comprises:

program code configured to compare the selected portions of the text data with a predefined list of homographs.

6. The computer program product of claim 1 wherein the program code for determining the preferred pronunciation comprises:

program code configured to modify the text data to indicate the preferred pronunciation of the identified homograph.

7. The computer program product of claim 6 wherein the program code configured to modify comprises:

program code configured to insert data defining the preferred pronunciation of the identified homograph into the text data.

8. The computer program product of claim 7 wherein the program code configured to insert comprises:

program code configured to substitute the identified homograph within the text data with data, comprehendible by the speech synthesizer, representing the preferred pronunciation of the identified homograph.

9. The computer program product of claim 1 wherein the program code which obtains the set of rules comprises program code which obtains an attribute table listing possible parts of speech for the identified homograph and a set of rules for each proposition pair of possible homograph parts of speech.

10. The computer program product of claim 9 wherein the set of rules are arranged in a predetermined order based on the identified homograph.

11. The computer program product of claim 10 wherein the program code which applies the rules applies the rules in the predetermined order.

12. The computer program product of claim 1 wherein the program code which determines a preferred pronunciation for the identified homograph retrieves the preferred pronunciation from a phonetic table.

13. A method for use with a computer system capable of converting text data into synthesized speech, the method comprising:

A. examining the text data to identify the homograph within the text data and to extract words surrounding the identified homograph in the text data;

B. using the identified homograph to identify the possible parts of speech that the identified homograph can assume;

C. using the possible parts of speech that the identified homograph can assume to obtain a set of rules, each rule based on a pair of possible parts of speech of the identified homograph and a word order and position of one of the surrounding words;

D. sequentially applying the rules in the obtained rule set until a rule is satisfied to determine a part of speech for the homograph in the text data; and

E. using the identified homograph and the determined part of speech usage for determining a preferred pronunciation for the identified homograph.

14. The method of claim 13 wherein step A comprises:

A.1 parsing the text data into phrases;

A.2 delineating the phrases by punctuation characters.

15. The method of claim 14 wherein step A2 further comprises:

A.2.1 comparing the parsed phrases with a predetermined list of punctuation characters.

16. The method of claim 13 wherein step A comprises:

A.1 parsing the text data into phrases; and

A.2 comparing the parsed phrases with a predetermined list of homographs.

17. The method of claim 13 wherein step D comprises:

D.1 modifying the text data to indicate the preferred pronunciation of the identified homograph.

18. The method of claim 17 wherein step D.1 further comprises the steps of:

D.1.1 inserting data, understandable by the speech synthesizer, representing the preferred pronunciation of the identified homograph; and

D.1.2 deleting the identified homograph from the text data.

19. The method of claim 13 wherein step B further comprises the steps of:

B.1 associating the identified homograph with an entry of an attribute table.

20. The method of claim 19 wherein step B further comprises the step of:

B.2 determining from the identified entry of the attribute table which grammatical function of language the homograph can perform.

21. The method of claim 20 wherein step B further comprises the step of:

B.3 performing a syntactic analysis of the identified homograph within the text.

22. The method of claim 21 wherein step B.3 further comprises the steps of:

B.3.1 analyzing the word order of the homograph within the text; and

B.3.2 analyzing the position of the homograph within the text.

23. The method of claim 20 wherein step B further comprises the step of:

B.3 performing the semantic analysis of the homograph within the text.

24. The method of claim 20 wherein step B further comprises the step of:

B.3 performing statistical analysis of the homograph within the text.

25. The method of claim 24 wherein step B.3 further comprises the step of:

B.3.1 determining from the identified entry for the homograph in the attribute table the preferred pronunciation from a statistics bit.

26. Apparatus for use with a computer system capable of converting text data into synthesized speech, the apparatus comprising:

a parser which examines the text data to identify the homograph within the text data and to extract words surrounding the identified homograph in the text data;

an attribute retriever responsive to the identified homograph which identifies the possible parts of speech that the identified homograph can assume;

a rules mechanism that uses the possible parts of speech that the identified homograph can assume and obtains a set of rules, each rule based on a pair of possible parts of speech of the identified homograph and a word order and position of one of the surrounding words;

a rules engine which sequentially applies the rules in the obtained rule set until a rule is satisfied to determine a part of speech for the homograph in the text data; and

a lookup mechanism which is responsive to the homograph and the determined part of speech usage for determining a preferred pronunciation for the identified homograph.

27. The apparatus of claim 26 wherein the attribute retriever comprises a mechanism which obtains an attribute table listing possible parts of speech for the identified homograph and a set of rules for proposition pairs of each possible homograph part of speech.

28. The apparatus of claim 27 wherein the set of rules are arranged in a predetermined order based on the identified homograph.

29. The apparatus of claim 28 wherein the rules engine applies the rules in the predetermined order.

30. The apparatus of claim 26 wherein the lookup mechanism retrieves the preferred pronunciation from a phonetic table.

31. A computer data signal embodied in a carrier wave for use with a computer system capable of converting text data into synthesized speech, the computer data signal comprising:

program code which examines the text data to identify the homograph within the text data and to extract words surrounding the identified homograph in the text data;

program code responsive to the identified homograph which identifies the possible parts of speech that the identified homograph can assume;

program code that uses the possible parts of speech that the identified homograph can assume and obtains a set of rules, each rule based on a possible pair of parts of speech of the identified homograph and a word order and position of one of the surrounding words;

program code which sequentially applies the rules in the obtained rule set until a rule is satisfied to determine a part of speech for the homograph in the text data; and

program code which is responsive to the homograph and the determined part of speech usage for determining a preferred pronunciation for the identified homograph.

32. The computer data signal of claim 31 wherein the program code which obtains the set of rules comprises program code which obtains an attribute table listing possible parts of speech for the identified homograph and a set of rules for each proposition pair of possible homograph parts of speech.

33. The computer data signal of claim 32 wherein the set of rules are arranged in a predetermined order based on the identified homograph.

34. The computer data signal of claim 33 wherein the program code which applies the rules applies the rules in the predetermined order.

35. The computer data signal of claim 31 wherein the program code which determines a preferred pronunciation for the identified homograph retrieves the preferred pronunciation from a phonetic table.