Method and apparatus for enhanced phonetic indexing and search

• US 9,311,914 B2
• Filed: 09/03/2012
• Issued: 04/12/2016
• Est. Priority Date: 09/03/2012
• Status: Active Grant

• Current Assignee: Nice Ltd
• Original Assignee: Nice Systems Limited (Nice Ltd)
• Inventors: Wasserblat, Moshe, Eylon, Dan, Ashkenazi, Tzach, Pereg, Oren, Laperdon, Ronen
• Primary Examiner(s): Shah, Paras D
• Assistant Examiner(s): THOMAS-HOMESCU, ANNE L
• Application Number:
  US13/602,214
• Publication Number:
  US 20140067373A1
• Time in Patent Office: 1,317 Days
• Field of Search: 707/741, 707/706, 707/780, 707/803, 707/769, 704/235, 704/238, 704/230, 704/260, 704/3, 704/254, 704/220, 704/231, 704/255, 704/251, 704/243, 704/234, 704/236, 704/270, 704/240, 715/753
• US Class Current: 1/1
• CPC Class Codes: G10L 15/02 Feature extraction for speech recognition; Selection of recognition unit
• CPC Class Codes: G10L 15/04 Segmentation; Word boundary detection
• CPC Class Codes: G10L 2015/025 Phonemes, fenemes or fenones being the recognition units

Abstract

The subject matter discloses a method two phase phonetic indexing and search comprising: receiving a digital representation of an audio signal; producing a phonetic index of the audio signal; producing phonetic N-gram sequence from the phonetic index by segmenting the phonetic index into a plurality of phonetic N-grams; and producing an inverted index of the plurality of phonetic N-grams.

Claims

1. A method comprising:

receiving digital representations of audio signals, wherein the audio signals are produced by a suitable device;

producing phonetic indexes of the audio signals based on said digital representations thereof;

producing phonetic N-gram sequences from the phonetic indexes by segmenting the phonetic indexes into a plurality of phonetic N-grams;

producing inverted indexes of the plurality of phonetic N-grams;

obtaining a textual search term;

converting the textual search term into a phonetic search term;

searching the phonetic search term on the inverted indexes;

ranking the digital representations of the audio signals based on matching between the search query and detected events of the search in the inverted indexes such that a rank reflects the probability that the textual search term is found within any of the audio signals,

wherein the ranking is according to the following formula;

2. The method according to claim 1, wherein the plurality of phonetic N-grams are of the same N-gram length.

3. The method according to claim 1, further comprises producing one or more segmented phonetic search terms from the phonetic search term, the segmented phonetic search terms having the same N-gram length as the plurality of phonetic N-grams;

producing a search query from the one or more segmented phonetic search terms; and

searching the search query on the inverted index.

4. The method according to claim 1, wherein the plurality of phonetic N-grams are of variable lengths.

5. The method according to claim 1, further comprises obtaining a limitation parameter;

and limiting the number of audio signals that are used for phonetic search according to the limitation parameter.

6. The method according to claim 1, further comprises determining whether to perform a phonetic search according to an accuracy estimation of the inverted index search.

7. The method according to claim 1, further comprises determining whether to perform, a phonetic search according to the available processing resources of a device performing the search.

8. The method according to claim 1, further comprises obtaining an enable/disable parameter;

and determining whether to perform a phonetic search according to the enable/disable parameter.

9. The method according to claim 1, further comprises storing search results of the inverted index in a storage device.

10. The method according to claim 1, wherein segmenting the phonetic index into a plurality of phonetic N-grams comprises segmenting the phonetic index to sequences of consecutive phonemes with an overlap of a whole number of phonemes between the sequences.

11. The method according to claim 10, wherein the whole number is zero, hence the plurality of phonetic N-grams do not overlap.

12. The method according to claim 10, wherein the whole number is non-zero, hence the plurality of phonetic N-grams overlap therebetween.

13. The method according to claim 1, wherein the second search would be performed only on audio signals that have a rank that is higher than a given threshold.

14. The method according to claim 1, wherein determining whether or not to perform the second search is according to an accuracy estimation of the search based on inverted indexes.

15. The method according to claim 1:

wherein determining whether or not to perform the second search is according to load balance of the computerized apparatus performing the phonetic search.

16. The method according to claim 1, wherein a is empirically determined.