User-aided adaptation of a phonetic dictionary

• US 9,922,643 B2
• Filed: 12/23/2014
• Issued: 03/20/2018
• Est. Priority Date: 12/23/2014
• Status: Active Grant

• Current Assignee: Nice Ltd
• Original Assignee: Nice Ltd
• Inventors: Nissan, Maor, Bretter, Ronny
• Primary Examiner(s): Dorvil, Richemond
• Assistant Examiner(s): Le, Thuykhanh
• Application Number:
  US14/580,331
• Publication Number:
  US 20160180835A1
• Time in Patent Office: 1,183 Days
• Field of Search: None
• 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
• CPC Class Codes: G10L 2015/025 Phonemes, fenemes or fenones being the recognition units
• CPC Class Codes: G10L 2015/0638 Interactive procedures

Abstract

A method for adapting a phonetic dictionary for peculiarities of a speech of an at least one speaker, comprising generating search pronunciations for a search term, retrieving audio sections from an audio database for each search pronunciation, audibly presenting to a person the audio sections of the speech of the at least one speaker, and updating the phonetic dictionary based on acceptability of the audio sections determined from judgments by the person regarding intelligibility of the audio sections in audibly pronouncing the provided at least one word, wherein the method is performed on an at least one computerized apparatus configured to perform the method.

Claims

1. A method for adapting a phonetic dictionary for peculiarities of a speech of at least one speaker, the method comprising:

a. receiving a search term in text form;

b. searching for the search term in the phonetic dictionary, wherein if the search term or a portion thereof is included in the phonetic dictionary,

obtaining from the phonetic dictionary pronunciation elements corresponding to phonetic elements of the search term or a portion thereof, for serving as search pronunciations of the search term or portion thereof,

and if the search term or portion thereof is not included in the phonetic dictionary,

generating pronunciation elements corresponding to phonetic elements of the search term or portion thereof with a text to phoneme model (TTP); and

concatenating all the obtained and generated pronunciation elements of the search term to form a search pronunciation of the search term;

c. searching by the formed search pronunciation for a matching phonetic transcription in an indexed transcriptions database, wherein the indexed transcriptions database stores phonetic transcriptions of a corresponding audio database of recorded human speech;

wherein the searching yields a found pronunciation;

d. retrieving an audio section corresponding to the matching phonetic transcription of the found pronunciation from the audio database;

e. audibly presenting to a person the audio section and tagging the audio section according to its acceptability for pronouncing of the search term to obtain a tagged search result;

f. calculating a value of acceptability efficiency level for the found pronunciation, as obtained from the indexed transcriptions database, wherein the value of acceptability determined by the tagging; and

g. in case the calculated value of acceptability exceeds a certain threshold value and the found pronunciation, as obtained from the indexed transcriptions database, is not already included in the phonetic dictionary, adding the found pronunciation to the phonetic dictionary;

wherein the method is performed on an at least one computerized apparatus configured to perform the method.

2. The method according to claim 1, wherein the acceptability of the pronunciation corresponding to a search term is calculated according to a confidence quantification for all tagged results that include the pronunciation and the corresponding search term, and wherein the phonetic dictionary is updated according to the confidence quantification.

3. The method according to the claim 2, wherein if a certain pronunciation and search term pair are determined acceptable according to the confidence quantification calculated for the tagged search results, and if the search term is included in the phonetic dictionary, then the phonetic dictionary is updated to include the certain pronunciation for the search term.

4. The method according to the claim 2, wherein if a certain pronunciation and search term pair is determined acceptable according to the confidence quantification calculated for the tagged search results, and if the search term is not included in the phonetic dictionary, then the phonetic dictionary is updated to include the search term and the certain pronunciation.

5. The method according to the claim 2, wherein if a certain pronunciation and search term pair are determined as unacceptable according to the confidence quantification calculated for the tagged search results, and if the search term and pronunciation are included in the phonetic dictionary, then the phonetic dictionary is updated by removing the certain pronunciation from the dictionary.

6. The method according to the claim 1, wherein the audio sections corresponding to the search pronunciations are retrieved from the speech of the at least one speaker in locations that correspond to the pronunciation elements of the search term, and wherein the retrieval is performed based on a metric that measures the distance between the search pronunciation and the found or recognized pronunciation elements.

7. The method according to the claim 1, wherein the search term is comprised of a plurality of words.

8. The method according to the claim 1, wherein the speech of the at least one speaker comprises speech of a plurality of speakers.

9. The method according to claim 1, wherein the method is further repeated at least one time, and the dictionary is updated at least one time.

10. The method according to claim 1, wherein the method is further repeated in order to acceptably update the phonetic dictionary according to an at least one provided criterion.

11. The method according to claim 1, further comprising obtaining or generating multiple search pronunciations for the search term, wherein the number of search pronunciations for the search term is equal to or exceeds a threshold number of pronunciations.

12. The method according to claim 11, wherein the audio sections are retrieved from the speech of the at least one speaker in locations that correspond to the pronunciation elements of the provided at least one search term.

13. The method according to claim 11, wherein the provided at least one search term comprises a plurality of provided search terms.

14. The method according to claim 11, wherein the speech of the at least one speaker comprises speech of a plurality of speakers.

15. The method according to claim 11, the method is repeated at least one time.

16. A computerized apparatus configured to carry out the method according to claim 1.