Diarization using acoustic labeling
Systems and method of diarization of audio files use an acoustic voiceprint model. A plurality of audio files are analyzed to arrive at an acoustic voiceprint model associated to an identified speaker. Metadata associate with an audio file is used to select an acoustic voiceprint model. The selected acoustic voiceprint model is applied in a diarization to identify audio data of the identified speaker.
1 . A method for diarization of audio files using acoustic voiceprints, the method comprising:
receiving an audio file for diarization by a processor, wherein the audio file includes at least two speakers speaking in the same audio file;
separating the audio file into a plurality of speaker segments, wherein each speaker segment is a segment of speech from at least one of the at least two speakers in the audio file, further wherein each segment is separated by a non-speech segment;
receiving a plurality of acoustic voiceprints from a voiceprint database server, wherein each acoustic voiceprint is trained from a plurality of speaker segments;
comparing the plurality of acoustic voiceprints to each speaker segment to determine which of the plurality of acoustic voiceprints match each speaker segment; and
applying a speaker label to each speaker segment based on the comparing, wherein the speaker label identifies a speaker associated with the speaker segment.
2 . The method of claim 1 , the method further comprising:
receiving speaker metadata for the audio file; and
selecting acoustic voiceprints from the plurality of acoustic voiceprints based upon the received speaker metadata for the comparing.
3 . The method of claim 1 , wherein each acoustic voice print is associated with a different known speaker.
4 . The method of claim 1 , wherein the audio file is real-time audio data of a customer service interaction including a customer service agent and at least one other speaker.
5 . The method of claim 1 , the method further comprising clustering similar speaker segments of the plurality of speaker segments, wherein the similar speaker segments have a high likelihood of containing speech from a single speaker.
6 . The method of claim 5 , wherein clustering the similar speaker segments of the plurality of speaker segments includes applying at least one metric to the speaker segments to label the segments of speech as belonging to a customer service agent or as belonging to an other speaker.
7 . The method of claim 6 , wherein the at least one metric is that of cluster size wherein the larger the cluster the more likely the segment belongs to the customer service agent.
8 . A system for diarization of audio files using acoustic voiceprints, the system comprising:
a memory comprising computer readable instructions;
a processor configured to read the computer readable instructions that when executed causes the system to:
receive an audio file for diarization by a processor, wherein the audio file includes at least two speakers speaking in the same audio file;
separate the audio file into a plurality of speaker segments, wherein each speaker segment is a segment of speech from at least one of the at least two speakers in the audio file, further wherein each segment is separated by a non-speech segment;
receive a plurality of acoustic voiceprints from a voiceprint database server, wherein each acoustic voiceprint is trained from a plurality of speaker segments;
compare the plurality of acoustic voiceprints to each speaker segment to determine which of the plurality of acoustic voiceprints match each speaker segment; and
apply a speaker label to each speaker segment based on the comparing, wherein the speaker label identifies a speaker associated with the speaker segment.
9 . The system of claim 8 , wherein the processor is further configured to cause the system to:
receive speaker metadata for the audio file; and
select acoustic voiceprints from the plurality of acoustic voiceprints based upon the received speaker metadata for the comparing.
10 . The system of claim 8 , wherein each acoustic voiceprint is associated with a different known speaker.
11 . The system of claim 8 , wherein the audio file is real-time audio data of a customer service interaction including a customer service agent and at least one other speaker.
12 . The system of claim 8 , wherein the processor is further configured to cause the system to: cluster similar speaker segments of the plurality of speaker segments, wherein the similar speaker segments have a high likelihood of containing speech from a single speaker.
13 . The system of claim 12 , wherein clustering the similar speaker segments of the plurality of speaker segments includes applying at least one metric to the speaker segments to label the segments of speech as belonging to a customer service agent or as belonging to an other speaker.
14 . The system of claim 13 , wherein the at least one metric is that of cluster size wherein the larger the cluster the more likely the segment belongs to the customer service agent.
15 . A non-transitory computer readable medium comprising computer readable code for diarization of audio files using acoustic voiceprints on a system that when executed by a processor, causes the system to:
receive an audio file for diarization by a processor, wherein the audio file includes at least two speakers speaking in the same audio file;
separate the audio file into a plurality of speaker segments, wherein each speaker segment is a segment of speech from at least one of the at least two speakers in the audio file, further wherein each segment is separated by a non-speech segment;
receive a plurality of acoustic voiceprints from a voiceprint database server, wherein each acoustic voiceprint is trained from a plurality of speaker segments;
compare the plurality of acoustic voiceprints to each speaker segment to determine which of the plurality of acoustic voiceprints match each speaker segment; and
apply a speaker label to each speaker segment based on the comparing, wherein the speaker label identifies a speaker associated with the speaker segment.
16 . The non-transitory computer readable medium of claim 15 , wherein the system is further caused to:
receive speaker metadata for the audio file; and
select acoustic voiceprints from the plurality of acoustic voiceprints based upon the received speaker metadata for the comparing.
17 . The non-transitory computer readable medium of claim 15 , wherein each acoustic voiceprint is associated with a different known speaker.
18 . The non-transitory computer readable medium of claim 15 , wherein the processor is further configured to cause the system to: cluster similar speaker segments of the plurality of speaker segments, wherein the similar speaker segments have a high likelihood of containing speech from a single speaker.
19 . The non-transitory computer readable medium of claim 18 , wherein clustering the similar speaker segments of the plurality of speaker segments includes applying at least one metric to the speaker segments to label the segments of speech as belonging to a customer service agent or as belonging to an other speaker.
20 . The non-transitory computer readable medium of claim 19 , wherein the at least one metric is that of cluster size wherein the larger the cluster the more likely the segment belongs to the customer service agent.