Method and apparatus for expanding the vocabulary of a speech system

• US 6,801,893 B1
• Filed: 06/22/2000
• Issued: 10/05/2004
• Est. Priority Date: 06/30/1999
• Status: Expired due to Term

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: International Business Machines Corporation
• Inventors: Crépy, Hubert, Backfried, Gerhard
• Primary Examiner(s): CHAWAN, VIJAY B
• Application Number:
  US09/599,129
• Time in Patent Office: 1,566 Days
• Field of Search: 704/257, 704/251, 704/270
• US Class Current: 704/257
• CPC Class Codes: G10L 15/063 Training

Abstract

A computerized method is provided for adding a new word to a vocabulary of a speech system, the vocabulary comprising words and corresponding acoustic patterns for a language or language domain. Within a determination step for the new word, a regularity value is determined which measures the conformity with respect to the pronunciation in the language or language domain. In a comparison step, the regularity value is compared to a threshold value to decide whether the conformity is insufficient. Only in the affirmative case of insufficient conformity, a prompting step is performed, prompting for additional information on the pronunciation of the new word. Finally, in an extension step, the new word and an acoustic pattern of the new word are added to the vocabulary.

Claims

1. A computer-based method of adding a new word to a vocabulary of a speech system, the vocabulary including words and corresponding acoustic patterns for a language or a language domain, the method comprising the steps of:

determining, for the new word, a regularity value measuring a conformity with respect to a pronunciation in the language or the language domain;

comparing the regularity value to a threshold value to determine a sufficiency of the conformity;

prompting for additional information on the pronunciation of the new word, when the conformity is insufficient; and

adding the new word and an acoustic pattern of the new word to the vocabulary.

2. The method of claim 1, wherein the regularity value is based on one of a computed perplexity and a computed log-probability of the spelling of the word with respect to a spelling language model.

3. The method of claim 2, wherein the language model is determined according to N-gram statistics.

4. The method of claim 1, wherein the additional information comprises a sounds-like-spelling pattern representing the pronunciation of the new word.

5. The method of claim 1, wherein the added acoustic pattern is derived from the additional information and from the pronunciation of the new word generated by a letter-to-sound system.

6. The method of claim 1, wherein the comparing step comprises the steps of:

determining regularity values for test words of a test corpus of words not part of the vocabulary;

determining a frequency distribution of the regularity values;

determining, from the frequency distribution, a reference regularity value as a relative maximum corresponding to the lowest conformity; and

choosing, as a threshold value, a value with a lower conformity than the reference regularity value.

7. The method of claim 1, wherein the determining step further comprises determining a multitude of regularity values with respect to a multitude of candidate vocabularies such that a selected vocabulary is chosen as the candidate vocabulary for which the regularity value indicates the largest conformity, and further wherein the added acoustic pattern is derived using the selected vocabulary.

8. The method of claim 1, wherein the speech system is one of a speech recognition system and a speech synthesis system.

9. Apparatus for adding a new word to a vocabulary of a speech system, the vocabulary including words and corresponding acoustic patterns for a language or a language domain, the apparatus comprising:

at least one processor operative to;

(i) determine, for the new word, a regularity value measuring a conformity with respect to a pronunciation in the language or the language domain;

(ii) compare the regularity value to a threshold value to determine a sufficiency of the conformity;

(iii) prompt for additional information on the pronunciation of the new word, when the conformity is insufficient; and

(iv) add the new word and an acoustic pattern of the new word to the vocabulary.

10. The apparatus of claim 9, wherein the regularity value is based on one of a computed perplexity and a computed log-probability of the spelling of the word with respect to a spelling language model.

11. The apparatus of claim 10, wherein the language model is determined according to N-gram statistics.

12. The apparatus of claim 9, wherein the additional information comprises a sounds-like-spelling pattern representing the pronunciation of the new word.

13. The apparatus of claim 9, wherein the added acoustic pattern is derived from the additional information and from the pronunciation of the new word generated by a letter-to-sound system.

14. The apparatus of claim 9, wherein the comparing operation comprises the steps of:

(i) determining regularity values for test words of a test corpus of words not part of the vocabulary;

(ii) determining a frequency distribution of the regularity values;

(iii) determining, from the frequency distribution, a reference regularity value as a relative maximum corresponding to the lowest conformity; and

(iv) choosing, as a threshold value, a value with a lower conformity than the reference regularity value.

15. The apparatus of claim 9, wherein the determining operation further comprises determining a multitude of regularity values with respect to a multitude of candidate vocabularies such that a selected vocabulary is chosen as the candidate vocabulary for which the regularity value indicates the largest conformity, and further wherein the added acoustic pattern is derived using the selected vocabulary.

16. The apparatus of claim 9, wherein the speech system is one of a speech recognition system and a speech synthesis system.

17. An article of manufacture for adding a new word to a vocabulary of a speech system, the vocabulary including words and corresponding acoustic patterns for a language or a language domain, comprising a machine readable medium containing one or more programs which when executed implement the steps of:

determining, for the new word, a regularity value measuring a conformity with respect to a pronunciation in the language or the language domain;

comparing the regularity value to a threshold value to determine a sufficiency of the conformity;

prompting for additional information on the pronunciation of the new word, when the conformity is insufficient; and

adding the new word and an acoustic pattern of the new word to the vocabulary.

18. The article of claim 17, wherein the regularity value is based on one of a computed perplexity and a computed log-probability of the spelling of the word with respect to a spelling language model.

19. The article of claim 18, wherein the language model is determined according to N-gram statistics.

20. The article of claim 17, wherein the additional information comprises a sounds-like-spelling pattern representing the pronunciation of the new word.

21. The article of claim 17, wherein the added acoustic pattern is derived from the additional information and from the pronunciation of the new word generated by a letter-to-sound system.

22. The article of claim 17, wherein the comparing step comprises the steps of:

determining regularity values for test words of a test corpus of words not part of the vocabulary;

determining a frequency distribution of the regularity values;

determining, from the frequency distribution, a reference regularity value as a relative maximum corresponding to the lowest conformity; and

choosing, as a threshold value, a value with a lower conformity than the reference regularity value.

23. The article of claim 17, wherein the determining step further comprises determining a multitude of regularity values with respect to a multitude of candidate vocabularies such that a selected vocabulary is chosen as the candidate vocabulary for which the regularity value indicates the largest conformity, and further wherein the added acoustic pattern is derived using the selected vocabulary.

24. The article of claim 17, wherein the speech system is one of a speech recognition system and a speech synthesis system.