Identical conversation detection systems and processes
Systems and methods are described relating to detection of audio of voice communications from different reference points containing an identical conversation or portion of a conversation. A method can use an analytics module operating on at least one processor. Call signatures of a plurality of audio of voice communications are received. An index loaded with the call signatures is created, where segments of the call signatures are associated with buckets of the index. A first subset of one or more of the buckets associated with a number of the segments that exceeds a threshold is identified. An identical conversation portion in two or more of the plurality of audio of voice communications is identified based on a second subset of one or more of the buckets, where the first subset of one or more of the buckets is disregarded.
1. A method comprising:
using an analytics module operating on at least one processor:
receiving call signatures for a plurality of voice communications;
associating a plurality of segments of the received call signatures with buckets of a hash table, wherein the call signature segments are associated with buckets based on a similarity to call signature segments already associated with the respective bucket;
identifying a first subset of the buckets that are associated with a number of the segments that exceeds a threshold number; and
identifying an identical conversation portion in two or more of the plurality of voice communications based on a second subset of the buckets that does not include the first subset buckets.
2. The method of claim 1 , wherein the call signatures are content independent.
3. The method of claim 1 , wherein the call signatures comprise sequences of seeds, each of the seeds comprising a peak frequency of a first vowel voiced in a respective one of the plurality of voice communications and a time difference between the first vowel and a second vowel voiced in the respective one of the plurality of voice communications.
4. The method of claim 1 , wherein the buckets are associated with segments of three seeds, each of the seeds comprising a peak frequency of a first vowel voiced in a respective one of the plurality of voice communications and a time difference between the first vowel and a second vowel voiced in the respective one of the plurality of voice communications.
5. The method of claim 1 , wherein the threshold number is large enough to render each of the buckets of the first subset to be statistically unlikely to indicate an identical conversation portion between respective call signatures associated with the respective bucket of the first subset.
6. The method of claim 1 further comprising, using a recording module operating on at least one processor, recording the voice communications.
7. The method of claim 1 further comprising, using a compression module operating on at least one processor, compressing the plurality of voice communications into the call signatures.
8. A system comprising:
one or more compression servers having one or more compression modules operable on at least one processor, the one or more compression modules being operable to generate call signatures from voice communications; and
an analytics server having an analytics module operable on at least one processor, the analytics module being operable to associate a plurality of segments of the call signatures with buckets of a hash table, wherein the call signature segments are associated with buckets based on a similarity to call signature segments already associated with the respective bucket, the analytics module further being operable to disregard ones of the buckets that are associated with a number of the call signatures that exceeds a threshold number, and operable to identify an identical conversation portion in two or more of the voice communications based on non-disregarded ones of the buckets.
9. The system of claim 8 further comprising:
a recording server having a recording module operable on at least one processor, the recording module being operable to store the voice communications to memory.
10. The system of claim 8 , wherein the call signatures comprise seeds, each of the seeds comprising a peak frequency of a first vowel voiced in a respective one of the audio of telephone calls and a time difference between the first vowel and a second vowel voiced in the respective one of the voice communications.
11. The system of claim 8 , wherein the buckets are segments of three seeds, each of the seeds comprising a peak frequency of a first vowel voiced in a respective one of the voice communications and a time difference between the first vowel and a second vowel voiced in the respective one of the voice communications.
12. The system of claim 8 , wherein the analytics server is at an analytics facility, and the one or more compression server comprises:
a first compression server on a first compression server device, the first compression server having a first compression module of the one or more compression modules operable on at least one processor, the first compression server device being located at a first controlled-environment facility remote from the analytics facility, and
a second compression server on a second compression server device, the second compression server having a second compression module of the one or more compression modules operable on at least one processor, the second compression server device being located at a second controlled-environment facility remote from the analytics facility, the second controlled-environment facility being remote from the first controlled-environment facility.
13. A non-transitory computer-readable storage medium having a computer program embodied thereon, the computer program comprising:
program code instructions for receiving call signatures for a plurality of voice communications;
program code instructions for associating a plurality of segments of the received call signatures with buckets of a hash table, wherein the call signature segments are associated with buckets based on a similarity to call signature segments already associated with the respective bucket;
program code instructions for identifying buckets that are associated with a number of call signatures that exceeds a threshold; and
program code instructions for determining whether two or more voice communications contain an identical conversation portion based on buckets that are associated with a number of call segments that does not exceed the threshold, wherein the buckets that are associated with the number of segments that exceeds the threshold are disregarded.
14. The non-transitory computer-readable storage medium of claim 13 , wherein the computer program comprises program code instructions for receiving the call signatures.
15. The non-transitory computer-readable storage medium of claim 13 , wherein the call signatures each comprise a sequence of seeds, each of the seeds comprising a peak frequency of a first vowel voiced in a respective one of the voice communications and a time difference between the first vowel and a second adjacent vowel voiced in the respective one of the voice communications.
16. The non-transitory computer-readable storage medium of claim 13 , wherein the buckets are segments of at least three seeds, each of the seeds comprising a peak frequency of a first vowel voiced in a respective one of the voice communications and a time difference between the first vowel and a second adjacent vowel voiced in the respective one of the voice communications.
17. The non-transitory computer-readable storage medium of claim 13 , wherein the threshold is twenty.