Detecting robocalls using biometric voice fingerprints

Abstract

The disclosed system and method detect robocalls using biometric voice fingerprints. The system receives audio input representing a plurality of telephone calls. For at least a portion of the telephone calls, the system analyzes the received audio based on a voice biometrics detection model to identify one or more biometric indicators characterizing a speaker in the analyzed telephone call. The system generates and stores a voice fingerprint characterizing the speaker based on the biometric indicators, and a time of the analyzed telephone call. The system analyzes stored voice fingerprints and times corresponding to speakers in the analyzed telephone calls to determine a frequency of occurrence of each voice fingerprint within an analyzed timeframe. If the frequency of occurrence of a voice fingerprint exceeds a threshold call quantity within the analyzed timeframe, the voice fingerprint is characterized as being associated with a robocaller.

1 - 20 . (canceled) 
     
     
         21 . A method performed by a computing system to identify a known caller in a received call using voice biometrics, the method comprising:
 receiving call audio for a call, the call audio containing real or simulated human speech of a speaker in the call audio;   generating, using a voice biometrics detection model, a biometric voice fingerprint for the speaker in the call audio,
 wherein the generated biometric voice fingerprint is based on multiple biometric indicators extracted from the call audio and is stored as a dimensional vector; 
   comparing the generated biometric voice fingerprint to at least some biometric voice fingerprints stored as dimensional vectors in a set of biometric voice fingerprints associated with known callers;   calculating a probability that the speaker in the call audio is a known caller based on a comparison between the generated biometric voice fingerprint and a biometric voice fingerprint in the set of biometric voice fingerprints; and   causing performance of an action depending on the calculated probability that the speaker in the call audio is the known caller,
 wherein the action includes allowing the call to proceed, generating an audio or visual warning associated with the call, generating a confirmation request to confirm an identity of the speaker, or terminating the call. 
   
     
     
         22 . The method of  claim 21 , wherein the multiple biometric indicators extracted from the call audio include at least one of volume, speaking rate, pitch, length of pauses, or duration of pauses. 
     
     
         23 . The method of  claim 21 , wherein the voice biometrics detection model is generated based on one or more artificial intelligence (AI) speech data processing models. 
     
     
         24 . The method of  claim 21 :
 wherein at least some of the known callers in the set of biometric voice fingerprints are each associated with a caller type, the caller type including a robocaller, a spam caller, or a legitimate caller, and   wherein calculating a probability that the speaker in the call audio is the known caller includes calculating a probability of a caller type for the speaker.   
     
     
         25 . The method of  claim 21 , wherein calculating a probability that the speaker in the call audio is a known caller includes calculating a similarity between the generated biometric voice fingerprint and the biometric voice fingerprint in the set of biometric voice fingerprints. 
     
     
         26 . The method of  claim 25 , wherein calculating a similarity comprises calculating a distance between the generated biometric voice fingerprint dimensional vector and the biometric voice fingerprint dimensional vectors in the set of biometric voice fingerprints. 
     
     
         27 . The method of  claim 21 , wherein the set of biometric voice fingerprints associated with the known callers includes at least one biometric voice fingerprint determined to be associated with a robocaller based on a frequency of occurrence of the at least one biometric voice fingerprint in a dataset comprising multiple voice fingerprints for callers detected in calls placed via a network during an analyzed timeframe. 
     
     
         28 . The method of  claim 21 , wherein the call audio includes a caller channel and a called channel, and wherein the multiple biometric indicators are extracted from the caller channel. 
     
     
         29 . The method of  claim 21 , wherein the audio or visual warning is a notification of the identification of the speaker. 
     
     
         30 . The method of  claim 21 , wherein the confirmation request is delivered via a graphical user interface (GUI), a text message, or an email. 
     
     
         31 . A non-transitory computer-readable medium carrying instructions that, when executed by a computing system, cause the computing system to perform operations to identify a known caller in a received call using voice biometrics, the operations comprising:
 receiving call audio for a call, the call audio containing real or simulated human speech of a speaker in the call audio;   generating, using a voice biometrics detection model, a biometric voice fingerprint for the speaker in the call audio,
 wherein the generated biometric voice fingerprint is based on multiple biometric indicators extracted from the call audio and is stored as a dimensional vector; 
   comparing the generated biometric voice fingerprint to at least some biometric voice fingerprints stored as dimensional vectors in a set of biometric voice fingerprints associated with known callers;   calculating a probability that the speaker in the call audio is a known caller based on a comparison between the generated biometric voice fingerprint and a biometric voice fingerprint in the set of biometric voice fingerprints; and   causing performance of an action depending on the calculated probability that the speaker in the call audio is the known caller.   
     
     
         32 . The non-transitory computer-readable medium of  claim 31 , wherein the action includes allowing the call to proceed. 
     
     
         33 . The non-transitory computer-readable medium of  claim 31 , wherein the action includes terminating the call, generating a confirmation request to confirm an identity of the speaker, or both. 
     
     
         34 . The non-transitory computer-readable medium of  claim 33 , wherein the confirmation request is delivered via a graphical user interface (GUI), a text message, or an email. 
     
     
         35 . The non-transitory computer-readable medium of  claim 31 , wherein the action includes generating an audio or visual warning associated with the call. 
     
     
         36 . The non-transitory computer-readable medium of  claim 31 , wherein the multiple biometric indicators extracted from the call audio include at least one of volume, speaking rate, pitch, length of pauses, or duration of pauses. 
     
     
         37 . The non-transitory computer-readable medium of  claim 31 , wherein the voice biometrics detection model is generated based on one or more artificial intelligence (AI) speech data processing models. 
     
     
         38 . The non-transitory computer-readable medium of  claim 31 , wherein calculating a probability that the speaker in the call audio is a known caller includes calculating a similarity between the generated biometric voice fingerprint and the biometric voice fingerprint in the set of biometric voice fingerprints. 
     
     
         39 . The non-transitory computer-readable medium of  claim 31 , wherein the set of biometric voice fingerprints associated with the known callers includes at least one biometric voice fingerprint determined to be associated with a robocaller based on a frequency of occurrence of the at least one biometric voice fingerprint in a dataset comprising multiple voice fingerprints for callers detected in calls placed via a network during an analyzed timeframe. 
     
     
         40 . The non-transitory computer-readable medium of  claim 31 , wherein the call audio includes a caller channel and a called channel, and wherein the multiple biometric indicators are extracted from the caller channel. 
     
     
         41 . The non-transitory computer-readable medium of  claim 31 :
 wherein at least some of the known callers in the set of biometric voice fingerprints are each associated with a caller type, the caller type including a robocaller, a spam caller, or a legitimate caller, and   wherein calculating a probability that the speaker in the call audio is the known caller includes calculating a probability of a caller type for the speaker.