Systems and methods for multi-stage language analysis and remediation of robocalls

A non-transitory computer-readable medium, storing a plurality of processor-executable instructions to:

• receive audio associated with a voice call having a particular duration;
• generate a waveform representation of the audio associated with the voice call;
• determine a proportion of the particular duration of the voice call in which an amplitude of the waveform representation of the audio associated with the voice call is above a threshold magnitude;
• determine, based on the proportion of the particular duration of the voice call in which the amplitude of the waveform representation of the audio is above the threshold magnitude, a first expected transcript length and a second expected transcript length, wherein the first expected transcript length is associated with a first language and the second expected transcript length is associated with a second language;
• generate or receive a first transcript, associated with the first language, of the voice call;
• generate or receive a second transcript, associated with the second language, of the voice call;
• identify a first length of the first transcript, associated with the first language, of the voice call;
• compare the first length of the first transcript to the first expected transcript length;
• identify a second length of the second transcript, associated with the second language, of the voice call;
• compare the second length of the second transcript to the first expected transcript length;
• determine, based on comparing the first length of the first transcript to the first expected transcript length and further based on comparing the second length of the second transcript to the second expected transcript length, that the voice call is associated with the first language;
• select a particular set of artificial intelligence/machine learning (“AI/ML”) models, out of a plurality of AI/ML models, based on determining that the voice call is associated with the first language;
• determine, based on the selected particular set of AI/ML models, that one or more remedial actions should be performed with respect to a caller associated with the voice call;
• and
• perform the one or more remedial actions, wherein the one or more remedial actions include blocking subsequent voice calls placed by the caller.

A method, comprising:

• receiving audio associated with a voice call having a particular duration;
• generating a waveform representation of the audio associated with the voice call;
• determining a proportion of the particular duration of the voice call in which an amplitude of the waveform representation of the audio associated with the voice call is above a threshold magnitude;
• determining, based on the proportion of the particular duration of the voice call in which the amplitude of the waveform representation of the audio is above the threshold magnitude, a first expected transcript length and a second expected transcript length, wherein the first expected transcript length is associated with a first language, and the second expected transcript length is associated with a second language;
• generating or receiving a first transcript, associated with the first language, of the voice call;
• generating or receiving a second transcript, associated with the second language, of the voice call;
• identifying a first length of the first transcript, associated with the first language, of the voice call;
• comparing the first length of the first transcript to the first expected transcript length;
• identifying a second length of the second transcript, associated with the second language, of the voice call;
• comparing the second length of the second transcript to the first expected transcript length;
• determining, based on comparing the first length of the first transcript to the first expected transcript length and further based on comparing the second length of the second transcript to the second expected transcript length, that the voice call is associated with the first language;
• selecting a particular set of artificial intelligence/machine learning (“AI/ML”) models, out of a plurality of AI/ML models, based on determining that the voice call is associated with the first language;
• determining, based on the selected particular set of AI/ML models, that one or more remedial actions should be performed with respect to a caller associated with the voice call;
• and
• performing the one or more remedial actions, wherein the one or more remedial actions include blocking subsequent voice calls placed by the caller.