System, method and apparatus for classifying communications in a communications system

Patent metadata

Field	Value
Publication number	US-9531873-B2
Application number	US-52142706-A
Country	US
Kind code	B2
Filing date	Sep 14, 2006
Priority date	Aug 13, 2004
Publication date	Dec 27, 2016
Grant date	Dec 27, 2016

Abstract

The present invention provides a system, method and apparatus for automatically classifying voice communications, such as voice messages and phone calls in prerecorded voicemails (one speaker) and two-way conversations, as either spam or legitimate signals in a communications system (e.g., SIP, IMS, UMA, etc.). More specifically, the present invention classifies a voice communication session by receiving one or more voice communication packets associated with the voice communication session, extracting one or more properties from the received voice communication packets and classifying the voice communication session based on the extracted properties. The present invention can also be implemented as a computer program embodied on a computer readable medium wherein each step is performed by one or more code segments.

First claim

What is claimed is:

1. A method for classifying a voice communication session comprising the steps of:
receiving, by a microprocessor, one or more voice communication packets associated with the voice communication session;
stripping off, by a microprocessor, one or more routing and transport protocols from the received voice communication packets to obtain one or more resulting voice data units;
accumulating, by a microprocessor, two or more of the resulting voice data units to obtain accumulated voice data;
converting, by a microprocessor, the accumulated voice data into at least one voice signal suitable for analysis by one or more classifiers, wherein each classifier uses one or more classification models;
providing, by a microprocessor, the at least one voice signal to one or more classifiers;
extracting one or more properties, by the one or more classifiers, from the at least one voice signal to obtain extracted properties;
classifying the voice communication session as a legitimate call or a spam call based on the extracted properties, wherein the classifying includes determining whether the at least one voice signal is at least one of machine-generated speech, commercial advertisements, and narrations; and
modifying the one or more classification models based on at least one of current operational data, historical operational data, user preferences, system preferences and feedback.

2. The method as recited in claim 1 , wherein the voice communication session complies with a SIP, an IMS or a UMA communications standard.

3. The method as recited in claim 1 , wherein the one or more properties are also extracted from a voice protocol data encoded within the received voice communication packets.

4. The method as recited in claim 1 , wherein the classification of the voice communication session further comprises a casual conversation, a commercial advertisement, a machine-generated speech, a scripted narration or an unknown call type.

5. The method as recited in claim 1 , further comprising the steps of:
assigning a unique identifier to the voice communication session; and
labeling the received voice communication packets with the assigned unique identifier.

6. The method as recited in claim 1 , further comprising the step of determining whether a certain amount of voice data has been extracted, accumulated and converted to classify the voice signal and delaying the step of classifying the voice communication session until the certain amount of voice data has been extracted, accumulated and converted.

7. The method as recited in claim 1 , further comprising the steps of:
extracting voice protocol data from the received voice communication packets; and
providing the voice protocol data to a codec classifier that extracts the one or more properties from the voice protocol data.

8. The method as recited in claim 1 , wherein the one or more classification models comprise:
a codec-based classification model, a wavelet classification model, an energy classification model, a pitch classification model or a recognition-based classification model; and
each classification model classifies the voice communication session based on one or more parameters.

9. The method as recited in claim 8 , further comprising the step of modifying the one or more parameters based on current operational data, historical operational data, user preferences, system preferences or feedback.

10. The method as recited in claim 1 , wherein the step of modifying the one or more classification models is automatically or manually executed.

11. The method as recited in claim 1 , wherein the step of modifying the one or more classification models is performed to correct one or more previous incorrect classifications.

12. The method as recited in claim 1 , further comprising the step of training the one or more classification models using the extracted properties.

13. The method as recited in claim 1 , wherein the one or more classification models comprise decision trees, neural networks or Bayesian networks.

14. The method as recited in claim 1 , further comprising the step of performing an action based on the classification of the voice communication session.

15. The method as recited in claim 14 , wherein the action comprises reporting the classification, providing validation data, taking no action, allowing the voice communication session, dropping the voice communication session, processing the voice communication session as spam or challenging an originator of the voice communication session.

16. A computation system, comprising:
a microprocessor and a computer-readable medium for classifying a voice communication session, wherein the computer-readable medium is coupled to the microprocessor, and wherein the microprocessor receives and executes instructions from the computer-readable medium and wherein the instructions cause the microprocessor to:
receive one or more voice communication packets associated with the voice communication session;
strip off one or more routing and transport protocols from the received voice communication packets to obtain one or more resulting voice data units;
accumulate two or more of the resulting voice data units to obtain accumulated voice data;
convert the accumulated voice data into at least one voice signal suitable for analysis by one or more classifiers;
provide the at least one voice signal to one or more classifiers, wherein each classifier extracts one or more properties from the at least one voice signal to obtain extracted properties;
classify the voice communication session as a legitimate call or a spam call based on the extracted properties, wherein the classifying includes determining whether the at least one voice signal is at least one of machine-generated speech, commercial advertisements, and narrations; and
modifying the one or more classification models based on at least one of current operational data, historical operational data, user preferences, system preferences and feedback.

17. The method as recited in claim 1 , wherein the extracted properties comprise at least one of energy and pitch characteristics of an uncompressed linear PCM form of the at least one voice signal, and a statistical property of a parameter generated by encoding using different standards.

18. A server coupled to a communication network and receiving input for classifying a voice communication session, wherein the server comprises a microprocessor and a computer-readable medium coupled thereto, the microprocessor receives instructions from the computer-readable medium and is programmed to:
Receive one or more voice communication packets associated with the voice communication session, strip off one or more routing and transport protocols from the received voice communication packets to obtain one or more resulting voice data units, accumulate two or more of the resulting voice data units to obtain accumulated voice data, convert the accumulated voice data into at least one voice signal suitable for analysis by one or more classifiers, and provide the at least one voice signal to the one or more classifiers, wherein each classifier extracts one or more properties from the at least one voice signal to obtain extracted properties, and classifies the voice communication session as a legitimate call or a spam call based on the extracted properties, wherein the classifying includes determining whether the at least one voice signal is at least one of machine-generated speech, commercial advertisements, and narrations, wherein each classifier uses

Assignees

• Kurapati Srikrishna
• Ayewah Nathaniel E
• Avaya Inc

Inventors

• Kurapati Srikrishna
• Ayewah Nathaniel E

Classifications

• H04L65/1079

  of unsolicited session attempts, e.g. SPIT · CPC title

• H04M3/436 Primary

  Arrangements for screening incoming calls {, i.e. evaluating the characteristics of a call before deciding whether to answer it (based on the calling party profile H04M3/42059; based on location H04M3/42348; based on presence H04M3/42365; diversion H04M3/54)} · CPC title

• H04M7/0078

  Security; Fraud detection; Fraud prevention · CPC title

• H04L63/1458

  Denial of Service · CPC title

• H04L12/585

  Electricity · mapped topic