Audio recording triage system

• US 9,697,825 B2
• Filed: 04/07/2015
• Issued: 07/04/2017
• Est. Priority Date: 04/07/2015
• Status: Active Grant

• Current Assignee: Nexidia Incorporated (Nice Ltd)
• Original Assignee: Nexidia Incorporated (Nice Ltd)
• Inventors: Arrowood, Jon A., Morris, Robert William, Reich, Mark
• Primary Examiner(s): REINIER, BARBARA DIANE
• Application Number:
  US14/680,138
• Publication Number:
  US 20160300565A1
• Time in Patent Office: 819 Days
• Field of Search: 704231, 704235, 704236, 704243, 704244, 704251
• CPC Class Codes: G10L 15/08 Speech classification or search
• CPC Class Codes: G10L 2015/088 Word spotting
• CPC Class Codes: G10L 25/27 characterised by the analysis technique

Abstract

A method for analysis of a set of media recordings includes, in a first processing stage, selecting a subset of the set of media recordings by performing a first processing of each media recording of the set of media recordings and in a second processing stage, performing a second processing of the subset of the set of media recordings, the second processing including extraction of information from the media recordings of the subset of the set of media recordings. The first processing of each media recording of the set of media recordings includes determining a characterization of the media recording, the characterization including at least a quantity characterizing a predicted performance of the second processing of the media recording.

Claims

1. A method for analysis of a set of media recordings, the method comprising:

in a first processing stage, selecting a subset of the set of media recordings by performing a first processing of each media recording of the set of media recordings; and

in a second processing stage, performing a second processing of the subset of the set of media recordings, the second processing including extraction of information from the media recordings of the subset of the set of media recordings;

wherein the first processing of each media recording of the set of media recordings includes determining a quantity characterizing a predicted performance of the second processing of the media recording, wherein the quantity represents a predicted extraction speed when extracting information from the media recording.

2. The method of claim 1 wherein each media recording of the set of media recordings includes an audio file.

3. The method of claim 1 wherein determining the predicted performance comprises:

determining a degree of presence of each acoustic pattern of a plurality of acoustic patterns in an audio content of the media recording; and

processing the determined degree of presence of each acoustic pattern of the plurality of acoustic patterns in the audio content of the media recording using a model for determining the characterization of the media recording based on the degree of presence of each acoustic pattern of a plurality of acoustic patterns in the audio content of the media recording.

4. The method of claim 3 wherein the model includes a linear support vector machine model.

5. The method of claim 3 where each acoustic pattern of the plurality of acoustic patterns includes a phoneme sequence.

6. The method of claim 3 wherein the degree of presence of each acoustic pattern of the plurality of acoustic patterns includes a number of occurrences of each acoustic pattern in the audio content of the media recording.

7. The method of claim 3 wherein the degree of presence of each acoustic pattern of the plurality of acoustic patterns includes a continuous time likelihood of occurrence of each acoustic pattern in audio content of the media recording.

8. The method of claim 1 wherein selecting the subset of the set of media recordings includes, for each media recording of the set of media recordings, applying a threshold to the quantity characterizing the predicted performance of the second processing of the media recording and selecting the media recording according to a result of applying the threshold.

9. The method of claim 1 wherein the quantity characterizing the predicted performance is based on a desired computation time.

10. The method of claim 1 wherein the quantity characterizing the predicted performance is based on a predicted automatic speech recognition accuracy.

11. The method of claim 1 wherein selecting the subset of the set of media recordings includes choosing a predetermined number of media recordings with the highest quantities characterizing the predicted performance of the second processing of the media recordings.

12. The method of claim 1 wherein the quantity characterizing the predicted performance of the second processing of the media recording additionally represents a predicted accuracy of information extraction.

13. The method of claim 1 wherein the quantity characterizing the predicted performance of the second processing of the media recording additionally represents a measure of a presence of one or more speakers in the media recording.

14. The method of claim 1 wherein the information extracted from the media recordings of the subset of the set of media recordings includes business metric information.

15. The method of claim 1 wherein the information extracted from the media recordings of the subset of the set of media recordings includes information related to a sentiment of one or more customers associated with one or more media recordings of the subset of the set of media recordings.

16. The method of claim 1 wherein the information extracted from the media recordings of the subset of the set of media recordings includes a classification of one or more media recordings of the subset of the set of media recordings into one or more media recording categories.

17. The method of claim 1 wherein the information extracted from the media recordings of the subset of the set of media recordings includes a measure of a performance of a customer service agent associated with one or more of the media recordings of the subset of the set of media recordings.

18. The method of claim 1 wherein the information extracted from the media recordings of the subset of the set of media recordings includes speech analytics information.

19. The method of claim 1, wherein the quantity additionally represents a desired computation time.

20. The method of claim 1, wherein the quantity additionally represents the predicted accuracy of automatic speech recognition (ASR).

21. The method of claim 1, wherein the quantity additionally represents a predicted recall for the media recording.

22. The method of claim 1, wherein in a first processing stage, the selected subset includes no more than a predetermined fixed number of media recordings.

23. Software stored in a non-transitory form on a computer-readable medium, for analysis of a set of media recordings, the software comprising instructions for causing a computing system to:

in a first processing stage, select a subset of the set of media recordings by performing a first processing of each media recording of the set of media recordings; and

in a second processing stage, perform a second processing of the subset of the set of media recordings, the second processing including extraction of information from the media recordings of the subset of the set of media recordings;

wherein the first processing of each media recording of the set of media recordings includes determining a quantity characterizing a predicted performance of the second processing of the media recording, wherein the quantity represents a predicted extraction speed when extracting information from the media recording.

24. A computing system for analysis of a set of media recordings, the computing system comprising:

at least one processor configured to process the set of media recordings, the processing including

in a first processing stage, selecting a subset of the set of media recordings by performing a first processing of each media recording of the set of media recordings; and

in a second processing stage, performing a second processing of the subset of the set of media recordings, the second processing including extraction of information from the media recordings of the subset of the set of media recordings;

wherein the first processing of each media recording of the set of media recordings includes determining a quantity characterizing a predicted performance of the second processing of the media recording, wherein the quantity represents a predicted extraction speed when extracting information from the media recording.

25. A method for analysis of a set of media recordings, the method comprising:

in a first processing stage, selecting a subset of the set of media recordings by performing a first processing of each media recording of the set of media recordings; and

in a second processing stage, performing a second processing of the subset of the set of media recordings, the second processing including extraction of information from the media recordings of the subset of the set of media recordings;

wherein the first processing of each media recording of the set of media recordings includes determining a characterization of the media recording, the characterization including at least a quantity characterizing a predicted performance of the second processing of the media recording, wherein the quantity characterizing the predicted performance of the second processing of the media recording represents a predicted information extraction speed.