Pronunciation discovery for spoken words

• US 8,577,681 B2
• Filed: 09/13/2004
• Issued: 11/05/2013
• Est. Priority Date: 09/11/2003
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Inventors: Roth, Daniel L., Gillick, Laurence S., Shire, Mike
• Primary Examiner(s): WOZNIAK, JAMES S
• Application Number:
  US10/939,942
• Publication Number:
  US 20050143970A1
• Time in Patent Office: 3,340 Days
• Field of Search: 704243-244, 704/251, 704/254
• US Class Current: 704/244
• CPC Class Codes: G10L 15/06 Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice G10L15/14 takes precedence
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 15/187 Phonemic context, e.g. pronunciation rules, phonotactical constraints or phoneme n-grams

Abstract

A method of generating an alternative pronunciation for a word or phrase, given an initial pronunciation and a spoken example of the word or phrase, includes providing the initial pronunciation of the word or phrase, and generating the alternative pronunciation by searching a neighborhood of pronunciations about the initial pronunciation via a constrained hypothesis, wherein the neighborhood includes pronunciations that differ from the initial pronunciation by at most one phoneme. The method further includes selecting a highest scoring pronunciation within the neighborhood of pronunciations.

Claims

1. A method comprising:

providing a lexicon;

receiving a spoken utterance of a word or phrase;

after receiving the spoken utterance, obtaining an initial pronunciation for the spoken utterance;

after obtaining the initial pronunciation, modifying the initial pronunciation to generate a plurality of alternative pronunciations;

using the received spoken utterance to score each of the pronunciations among the plurality of alternative pronunciations;

identifying a highest scoring pronunciation among the plurality of alternative pronunciations; and

updating the lexicon with the highest scoring pronunciation.

2. The method of claim 1, wherein the plurality of alternative pronunciations differ from the initial pronunciation in a limited number of phonemes.

3. The method of claim 1, wherein the plurality of alternative pronunciations differ from the initial pronunciation by at most one phoneme.

4. The method of claim 1, wherein obtaining the initial pronunciation involves accepting a recognized transcript from a speech recognizer.

5. The method of claim 1, wherein obtaining the initial pronunciation involves accepting a user selection of a recognized transcript.

6. The method of claim 1, wherein obtaining the initial pronunciation involves accepting a user selection from a list of alternatives to a recognized transcript.

7. The method of claim 1, further comprising using a phonetic recognizer to compute a score for each of the alternative pronunciation and/or the initial pronunciation, and using one or both of these scores to decide whether to add the alternative pronunciation to a lexicon.

8. The method of claim 1, wherein updating a lexicon by replacing another pronunciation with the highest-scoring alternative pronunciation.

9. The method of claim 8, further comprising providing a user with an option of allowing or disallowing the update of the lexicon.

10. The method of claim 1, wherein the lexicon is a phonetic lexicon.

11. The method of claim 10, further comprising:

using the phonetic lexicon to perform recognition on the spoken utterance to generate a recognized transcript of the spoken utterance, said recognized transcript being characterized by a corresponding pronunciation; and

generating an N-best list of alternatives to the recognized transcript, each alternative of that N-best list of alternatives having a corresponding pronunciation,

wherein the initial pronunciation is selected from the N-best list of alternatives.

12. The method of claim 11, wherein said recognized transcript has a score and wherein updating the lexicon involves first determining that the highest scoring pronunciation among the plurality of alternative pronunciations has a score that is higher than the score of the pronunciation for the recognized transcript and then adding the highest scoring pronunciation to the lexicon.

13. The method of claim 12, wherein updating the lexicon also involves deleting said corresponding pronunciation for the recognized transcript.

14. The method of claim 1, further comprising using the lexicon to perform recognition on the spoken utterance to generate a recognized transcript of the spoken utterance, said recognized transcript being characterized by a corresponding pronunciation, and wherein the initial pronunciation is different from said corresponding pronunciation for the recognized transcript.

15. The method of claim 7, wherein the lexicon is a phonetic lexicon.

16. The method of claim 15, further comprising:

using the phonetic lexicon to perform recognition on the spoken utterance to generate a recognized transcript of the spoken utterance, said recognized transcript being characterized by a corresponding pronunciation; and

generating an N-best list of alternatives to the recognized transcript, each alternative of that N-best list of alternatives having a corresponding pronunciation,

wherein the initial pronunciation is selected from the N-best list of alternatives.

17. The method of claim 16, wherein said recognized transcript has a score and wherein updating the lexicon involves first determining that the highest scoring pronunciation among the plurality of alternative pronunciations has a score that is higher than the score of the pronunciation for the recognized transcript and then adding the highest scoring pronunciation to the lexicon.

18. The method of claim 17, wherein updating the lexicon also involves deleting said corresponding pronunciation for the recognized transcript.

19. The method of claim 7, further comprising using the lexicon to perform recognition on the spoken utterance to generate a recognized transcript of the spoken utterance, said recognized transcript being characterized by a corresponding pronunciation, and wherein the initial pronunciation is different from said corresponding pronunciation for the recognized transcript.

20. A method comprising:

providing a lexicon;

receiving a spoken utterance of a word or phrase;

after receiving the spoken utterance, obtaining an initial pronunciation for the spoken utterance;

after obtaining the initial pronunciation, modifying the initial pronunciation to generate a plurality of alternative pronunciations by changing the initial pronunciation by one phoneme;

identifying a highest scoring pronunciation among the plurality of alternative pronunciations; and

updating the lexicon with the highest scoring pronunciation.

21. A non-transitory computer readable medium storing executable instructions which when executed on a computer system cause the computer system to:

receive a spoken utterance of a word or phrase;

after receiving the spoken utterance, obtain an initial pronunciation for the spoken utterance;

after obtaining the initial pronunciation, modify the initial pronunciation to generate a plurality of alternative pronunciations by changing the initial pronunciation by one phoneme; and

use the received spoken utterance to score each of the pronunciations among the plurality of alternative pronunciations;

identify a highest scoring pronunciation among the plurality of alternative pronunciations; and

update a lexicon with the highest scoring pronunciation.

22. The computer readable medium of claim 21, wherein the stored executable instructions when executed on the computer system cause the computer system to obtain the initial pronunciation by accepting a user selection of a recognized transcript.

23. The computer readable medium of claim 21, wherein the stored executable instructions when executed on the computer system also cause the computer system to present a user with a list of alternatives to a recognized transcript and to accept a user selection from the list of alternatives as the initial pronunciation.

24. The computer readable medium of claim 21, wherein the stored executable instructions when executed on the computer system also cause the computer system to implement a phonetic recognizer which computes a score for each of the alternative pronunciations.

25. The computer readable medium of claim 21, wherein the stored executable instructions when executed on the computer system also cause the computer system to update the lexicon by replacing a stored phonetic representation with the highest-scoring potential alternative pronunciation.

26. The computer readable medium of claim 21, wherein the lexicon is a phonetic lexicon.

27. The computer readable medium of claim 26, wherein the stored executable instructions when executed on the computer system also cause the computer system to:

use the phonetic lexicon to perform recognition on the spoken utterance to generate a recognized transcript of the spoken utterance, said recognized transcript being characterized by a corresponding pronunciation; and

generate an N-best list of alternatives to the recognized transcript, each alternative of that N-best list of alternatives having a corresponding pronunciation,

wherein the initial pronunciation is selected from the N-best list of alternatives.

28. The computer readable medium of claim 27, wherein said recognized transcript has a score and wherein the stored executable instructions when executed on the computer system also cause the computer system to update the lexicon by first determining that the highest scoring pronunciation among the plurality of alternative pronunciations has a score that is higher than the score of the pronunciation for the recognized transcript and then adding the highest scoring pronunciation to the lexicon.

29. The computer readable medium of claim 28, wherein the stored executable instructions when executed on the computer system also cause the computer system to update the lexicon by deleting said corresponding pronunciation for the recognized transcript.

30. The computer readable medium of claim 21, wherein the stored executable instructions when executed on the computer system also cause the computer system to use the lexicon to perform recognition on the spoken utterance to generate a recognized transcript of the spoken utterance, said recognized transcript being characterized by a corresponding pronunciation, and wherein the initial pronunciation is different from said corresponding pronunciation for the recognized transcript.