Analysis and matching of voice signals

1. A fraud detection system comprising:

a processor and a computer readable medium operably coupled thereto, the computer readable medium comprising a plurality of instructions stored in association therewith that are accessible to, and executable by, the processor, to perform operations which comprise:

receiving a plurality of call interactions;

extracting a voice print of a caller from each of the call interactions;

determining which call interactions are associated with a single caller by comparing and matching pairs of voice prints of the call interactions;

organizing the call interactions associated with a single caller into a group; and

determining that a matching phrase was spoken by the single caller in a first call interaction and a second call interaction in the group, wherein determining that a matching phrase was spoken by the single caller in a first call interaction and a second call interaction in the group comprises:

converting a speech segment in the first call interaction and a speech segment in the second call interaction into feature vectors for each frame of speech;

calculating a distance between each frame of speech in the first call interaction and each frame of speech in the second call interaction;

storing the calculated distances in a matrix; and

starting at each frame of speech in the first call interaction and each frame of speech in the second call interaction, computing, with a sliding dynamic time warping algorithm, a warping distance.

2. The fraud detection system of claim 1 , wherein the operations further comprise generating an audio file, wherein the audio file comprises the single caller speaking the matching phrase in the first call interaction and the single caller speaking the matching phrase in the second call interaction.

3. The fraud detection system of claim 2 , wherein the operations further comprise presenting the audio file to a user.

4. The fraud detection system of claim 2 , wherein the operations further comprise:

determining that the matching phrase was spoken by the single caller in a third call interaction; and

adding the single caller speaking the matching phrase in the third call interaction to the audio file.

5. The fraud detection system of claim 1 , wherein the operations further comprise scoring each computation based on its warping distance, wherein a lower warping distance is given a higher score.

6. The fraud detection system of claim 5 , wherein the operations further comprise presenting a score for the computation, and start and end times for the matching phrase in the first call interaction and the second call interaction.

7. The fraud detection system of claim 1 , wherein determining that a matching phrase was spoken by the single caller in a first call interaction and a second call interaction in the group further comprises dividing each of the first call interaction and the second call interaction into a plurality of speech segments.

8. The fraud detection system of claim 1 , wherein the operations further comprise filtering the plurality of call interactions to identify high-risk call interactions.

9. The fraud detection system of claim 1 , wherein the operations further comprise presenting the first call interaction and the second call interaction to a user.

10. A method of detecting fraud, which comprises:

receiving a plurality of call interactions;

extracting a voice print of a caller from each of the call interactions;

determining which call interactions are associated with a single caller by comparing and matching pairs of voice prints of the call interactions;

organizing the call interactions associated with a single caller into a group; and

determining that a matching phrase was spoken by the single caller in a first call interaction and a second call interaction in the group, wherein determining that a matching phrase was spoken by the single caller in a first call interaction and a second call interaction in the group comprises:

converting a speech segment in the first call interaction and a speech segment in the second call interaction into feature vectors for each frame of speech;

calculating a distance between each frame of speech in the first call interaction and each frame of speech in the second call interaction;

storing the calculated distances in a matrix; and

starting at each frame of speech in the first call interaction and each frame of speech in the second call interaction, computing, with a sliding dynamic time warping algorithm, a warping distance.

11. The method of claim 10 , which further comprises:

generating an audio file, wherein the audio file comprises the single caller speaking the matching phrase in the first call interaction and the single caller speaking the matching phrase in the second call interaction; and

presenting the audio file to a user.

12. The method of claim 11 , which further comprises receiving verification from the user that the single caller is one person.

13. The method of claim 10 , which further comprises presenting the first call interaction and the second call interaction to a user.

14. A non-transitory computer-readable medium having stored thereon computer-readable instructions executable by a processor to perform operations which comprise:

receiving a plurality of call interactions;

extracting a voice print of a caller from each of the call interactions;

determining which call interactions are associated with a single caller by comparing and matching pairs of voice prints of the call interactions;

organizing the call interactions associated with a single caller into a group; and

determining that a matching phrase was spoken by the single caller in a first call interaction and a second call interaction in the group, wherein determining that a matching phrase was spoken by the single caller in a first call interaction and a second call interaction in the group comprises:

converting a speech segment in the first call interaction and a speech segment in the second call interaction into feature vectors for each frame of speech;

calculating a distance between each frame of speech in the first call interaction and each frame of speech in the second call interaction;

storing the calculated distances in a matrix; and

starting at each frame of speech in the first call interaction and each frame of speech in the second call interaction, computing, with a sliding dynamic time warping algorithm, a warping distance.

15. The non-transitory computer-readable medium of claim 14 , wherein the operations further comprise:

generating an audio file, wherein the audio file comprises the single caller speaking the matching phrase in the first call interaction and the single caller speaking the matching phrase in the second call interaction; and

presenting the audio file to a user.

16. The non-transitory computer-readable medium of claim 15 , wherein the operations further comprise:

determining that the matching phrase was spoken by the single caller in a third call interaction; and

adding the single caller speaking the matching phrase in the third call interaction to the audio file.

17. The non-transitory computer-readable medium of claim 14 , wherein the operations further comprise presenting the first call interaction and the second call interaction to a user.