Compact text-to-phone pronunciation dictionary

• US 7,080,005 B1
• Filed: 06/08/2000
• Issued: 07/18/2006
• Est. Priority Date: 07/19/1999
• Status: Active Grant

• Current Assignee: Texas Instruments, Inc.
• Original Assignee: Texas Instruments, Inc.
• Inventors: Kao, Yu-Hung
• Primary Examiner(s): ARMSTRONG, ANGELA A
• Application Number:
  US09/590,613
• Time in Patent Office: 2,231 Days
• Field of Search: 704/7, 704/10, 704/243, 704/244, 704/254, 704/9
• US Class Current: 704/10
• CPC Class Codes: G06F 40/242 Dictionaries
• CPC Class Codes: G10L 15/187 Phonemic context, e.g. pronunciation rules, phonotactical constraints or phoneme n-grams

Abstract

A typical English pronunciation dictionary takes up to 1,826,302 bytes in ASCII to store. A five times compression while maintaining computability is achieved by prefix delta encoding of the word and error encoding of the pronunciation.

Claims

1. A processor for creating a reduced size encoded pronunciation dictionary from an input pronunciation dictionary such that the encoded pronunciation dictionary does not need to be expanded to a larger size in order to be utilized, comprising:

a reading and sorting processor to read an input pronunciation dictionary and sort the words of the dictionary in alphabetical order;

a word encoder that encodes each word of the pronunciation dictionary by comparing the word with the prior word encoded and either outputs the number of prefix characters that match both the word and the prior word beginning characters if the number of matching characters is greater than or equal to N followed by the suffix characters of the word after character N, or outputs all characters of the word if the number of prefix characters that match the word and prior encoded word is less than N;

a text-to-pronunciation processor that operates on the word to be encoded to generate a pronunciation hypothesis;

a pronunciation comparer that compares the pronunciation of each word by comparing the pronunciation of the word from the input pronunciation dictionary and the pronunciation hypothesis from the text-to-pronunciation processor and determines the minimum number of pronunciation differences consisting of substitutions, deletions and insertions that need to be corrected in the pronunciation hypothesis to convert it to match the pronunciation of the input pronunciation dictionary; and

a pronunciation encoder that compares the pronunciation differences of the word to the pronunciation differences of the prior encoded word and either outputs the number of prefix differences that match the beginning of both the word differences and the prior word differences followed by the suffix differences of the word if the number of prefix matching differences is greater than or equal to M, or outputs all differences to the word if the number of prefix differences that match the word and prior encoded word is less than N.

2. The processor of claim 1 where N=2.

3. The processor of claim 1 where M=1.

4. The processor of claim 1 where the input pronunciation dictionary and the text-to-pronunciation processor provide the pronunciation of a word in terms of phones of a language.

5. The processor of claim 1 where the output characters of the word encoder are in bytes.

6. The processor of claim 1 where the pronunciation encoder outputs the number of prefix differences that match as a byte character.

7. The processor of claim 1 where the pronunciation encoder outputs a substitution difference as two byte characters which indicate the difference is a substitution, the location of the difference, and the pronunciation difference to substitute.

8. The processor of claim 1 where the pronunciation encoder outputs a deletion difference as two byte characters which indicate the difference is a deletion, the location of the difference, and the pronunciation difference that was deleted.

9. The processor of claim 1 where the pronunciation encoder outputs an insertion difference as one byte character which indicates the difference is an insertion and the location of the difference.

10. A method for creating a reduced size encoded pronunciation dictionary from an input pronunciation dictionary such that the encoded pronunciation dictionary does not need to be expanded to a larger size in order to be utilized, comprising the steps of:

reading an input pronunciation dictionary and sorting the words of the dictionary in alphabetical order using a processor;

encoding each word of the pronunciation dictionary by comparing the word with the prior word encoded and either outputting the number of prefix characters that match both the word and the prior word beginning characters if the number of matching characters is greater than or equal to N followed by the suffix characters of the word after character N, or outputting all characters of the word if the number of prefix characters that match the word and prior encoded word is less than N;

operating on the word to be encoded by a text-to-pronunciation processor to generate a pronunciation hypothesis;

comparing the pronunciation of each word using a pronunciation comparer by comparing the pronunciation of the word from the input pronunciation dictionary and the pronunciation hypothesis from the text-to-pronunciation processor and determining the minimum number of pronunciation differences consisting of substitutions, deletions and insertions that need to be corrected in the pronunciation hypothesis to convert it to match the pronunciation of the input pronunciation dictionary; and

comparing the pronunciation differences of the word to the pronunciation differences of the prior encoded word by a pronunciation encoder and either outputting the number of prefix differences that match the beginning of both the word differences and the prior word differences followed by the suffix differences of the word if the number of prefix matching differences is greater than or equal to M, or outputting all differences to the word if the number of prefix differences that match the word and prior encoded word is less than N.

11. The method of claim 10 where N=2.

12. The method of claim 10 where M=1.

13. The method of claim 10 where the input pronunciation dictionary and the text-to-pronunciation processor provide the pronunciation of a word in terms of phones of a language.

14. The method of claim 10 where the output characters of the word encoder are in bytes.

15. The method of claim 10 where the pronunciation encoder outputs the number of prefix differences that match as a byte character.

16. The method of claim 10 where the pronunciation encoder outputs a substitution difference as two byte characters which indicate the difference is a substitution, the location of the difference, and the pronunciation difference to substitute.

17. The method of claim 10 where the pronunciation encoder outputs a deletion difference as two byte characters which indicate the difference is a deletion, the location of the difference, and the pronunciation difference that was deleted.

18. The method of claim 10 where the pronunciation encoder outputs an insertion difference as one byte character which indicates the difference is an insertion and the location of the difference.