Voicemail spam detection
Voicemail spam detection is performed based on content of voicemail messages. The content of an incoming voicemail message is compared to a spam template that includes a representation of a spam voicemail. Spam templates may be generated based on spam indications provided by users for voicemail messages they have received. User indications for sufficiently similar voicemail messages may be aggregated by maintaining a vote count for a spam template that reflects how many times a user has indicated a matching voicemail message is spam. A spam template may also include an occurrence count that reflects how many times voicemail messages matching a spam template have been detected in a telephony system. An incoming voicemail message may be compared to spam templates and, responsive to a match of content and/or a corresponding vote count or occurrence count meeting a condition, the voicemail message may be identified as spam.
1. A method, comprising:
receiving a first voicemail message;
comparing the first voicemail message to a voicemail representation of a spam template, wherein the spam template includes a vote count that reflects a number of times that one or more users have indicated that a voicemail message matching the voicemail representation is spam, and wherein the voicemail representation of the spam template includes one or more word vectors, and comparing the first voicemail message to the voicemail representation of the spam template comprises:
applying a speech recognition module to the first voicemail message to obtain a text transcript representing the first voicemail message,
inputting the text transcript to a language model to obtain one or more word vectors representing the first voicemail message, and
comparing the one or more word vectors representing the first voicemail message to the one or more word vectors of the spam template, wherein comparing the one or more word vectors representing the first voicemail message to the one or more word vectors of the spam template comprises:
determining a cosine similarity between word vectors;
comparing the vote count of the spam template to a first threshold;
responsive to a match between the first voicemail message and the voicemail representation of the spam template and to the vote count exceeding the first threshold, identifying the first voicemail message as spam; and
responsive to identifying the first voicemail message as spam, invoking a spam mitigation measure on the first voicemail message.
2. The method of claim 1 , wherein the voicemail representation of the spam template includes an audio recording, and comparing the first voicemail message to the voicemail representation of the spam template comprises:
determining a distance metric between the audio recording of the spam template and an audio recording of the first voicemail message.
3. The method of claim 1 , wherein the spam template further comprises an occurrence count reflecting a number of received voicemail messages that have been matched to the voicemail representation of the spam template, and further comprising:
comparing the occurrence count of the spam template to a second threshold, wherein the first voicemail message is identified as spam responsive to the occurrence count exceeding the second threshold.
4. The method of claim 1 , further comprising:
receiving a second voicemail message;
receiving an indication from a user that the second voicemail message is spam;
comparing the second voicemail message to the voicemail representation of the spam template; and
responsive to a match between the second voicemail message and the voicemail representation of the spam template, incrementing the vote count of the spam template.
5. The method of claim 1 , wherein the spam mitigation measure includes blocking the first voicemail message from entering a voicemail inbox of a user.
6. The method of claim 1 , wherein the spam mitigation measure includes deleting the first voicemail message.
7. The method of claim 1 , wherein the spam mitigation measure includes displaying a message in a user interface indicating that the first voicemail message is spam.
8. A system, comprising:
a processor, and
a memory, wherein the memory stores instructions executable by the processor to:
receive a first voicemail message;
compare the first voicemail message to a voicemail representation of a spam template, wherein the spam template includes a vote count that reflects a number of times that one or more users have indicated that a voicemail message matching the voicemail representation is spam, and wherein the voicemail representation of the spam template includes one or more word vectors and the memory stores instructions executable by the processor to:
apply a speech recognition module to the first voicemail message to obtain a text transcript representing the first voicemail message;
input the text transcript to a language model to obtain one or more word vectors representing the first voicemail message; and
compare the one or more word vectors representing the first voicemail message to the one or more word vectors of the spam template, wherein the memory stores instructions executable by the processor to:
determine a cosine similarity between word vectors;
compare the vote count of the spam template to a first threshold; and
responsive to a match between the first voicemail message and the voicemail representation of the spam template and to the vote count exceeding the first threshold, identify the first voicemail message as spam; and
responsive to identifying the first voicemail message as spam, invoke a spam mitigation measure on the first voicemail message.
9. The system of claim 8 , wherein the voicemail representation of the spam template includes an audio recording, and the memory stores instructions executable by the processor to:
determine a distance metric between the audio recording of the spam template and an audio recording of the first voicemail message.
10. The system of claim 8 , wherein the spam template further comprises an occurrence count reflecting a number of received voicemail messages that have been matched to the voicemail representation of the spam template, and the memory stores instructions executable by the processor to:
compare the occurrence count of the spam template to a second threshold, wherein the first voicemail message is identified as spam responsive to the occurrence count exceeding the second threshold.
11. The system of claim 8 , wherein the memory stores instructions executable by the processor to:
receive a second voicemail message;
receive an indication from a user that the second voicemail message is spam;
compare the second voicemail message to the voicemail representation of the spam template; and
responsive to a match between the second voicemail message and the voicemail representation of the spam template, increment the vote count of the spam template.
12. The system of claim 8 , wherein the spam mitigation measure includes blocking the first voicemail message from entering a voicemail inbox of a user.
13. The system of claim 8 , wherein the spam mitigation measure includes displaying a message in a user interface indicating that the first voicemail message is spam.
14. A non-transitory computer-readable storage medium, comprising executable instructions that, when executed by a processor, facilitate performance of operations, comprising:
receiving a first voicemail message;
comparing the first voicemail message to a voicemail representation of a spam template, wherein the spam template includes a vote count that reflects a number of times that one or more users have indicated that a voicemail message matching the voicemail representation is spam, and wherein the voicemail representation of the spam template includes one or more word vectors, and comparing the first voicemail message to the voicemail representation of the spam template comprises:
applying a speech recognition module to the first voicemail message to obtain a text transcript representing the first voicemail message,
inputting the text transcript to a language model to obtain one or more word vectors representing the first voicemail message, and
comparing the one or more word vectors representing the first voicemail message to the one or more word vectors of the spam template, wherein comparing the one or more word vectors representing the first voicemail message to the one or more word vectors of the spam template comprises:
determining a cosine similarity between word vectors;
comparing the vote count of the spam template to a first threshold;
responsive to a match between the first voicemail message and the voicemail representation of the spam template and to the vote count exceeding the first threshold, identifying the first voicemail message as spam; and
responsive to identifying the first voicemail message as spam, invoking a spam mitigation measure on the first voicemail message.
15. The non-transitory computer-readable storage medium of claim 14 , wherein the voicemail representation of the spam template includes an audio recording, and comparing the first voicemail message to the voicemail representation of the spam template comprises:
determining a distance metric between the audio recording of the spam template and an audio recording of the first voicemail message.
16. The non-transitory computer-readable storage medium of claim 14 , wherein the spam template further comprises an occurrence count reflecting a number of received voicemail messages that have been matched to the voicemail representation of the spam template, and the operations further comprise:
comparing the occurrence count of the spam template to a second threshold, wherein the first voicemail message is identified as spam responsive to the occurrence count exceeding the second threshold.
17. The non-transitory computer-readable storage medium of claim 14 , wherein the operations further comprise:
receiving a second voicemail message;
receiving an indication from a user that the second voicemail message is spam;
comparing the second voicemail message to the voicemail representation of the spam template; and
responsive to a match between the second voicemail message and the voicemail representation of the spam template, incrementing the vote count of the spam template.
18. The non-transitory computer-readable storage medium of claim 14 , wherein the spam mitigation measure includes blocking the first voicemail message from entering a voicemail inbox of a user.
19. The non-transitory computer-readable storage medium of claim 14 , wherein the spam mitigation measure includes deleting the first voicemail message.
20. The non-transitory computer-readable storage medium of claim 14 , wherein the spam mitigation measure includes displaying a message in a user interface indicating that the first voicemail message is spam.