Modified media detection

Patent metadata

Field	Value
Publication number	US-11514342-B2
Application number	US-202016915888-A
Country	US
Kind code	B2
Filing date	Jun 29, 2020
Priority date	Jun 29, 2020
Publication date	Nov 29, 2022
Grant date	Nov 29, 2022

Abstract

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for detecting modified media are disclosed. In one aspect, a method includes the actions of receiving an item of media content. The actions further include providing the item as an input to a model that is configured to determine whether the item likely includes audio of a user's voice that was not spoken by the user or likely includes video of the user that depicts actions of the user that were not performed by the user. The actions further include receiving, from the model, data indicating whether the item likely includes audio of the user's voice that was not spoken by the user or includes video of the user that depicts actions of the user that were not performed by the user. The actions further include determining whether the item likely includes deepfake content.

First claim

What is claimed is:

1. A computer-implemented method comprising:
receiving, by a computing device, media data that represents an item of media content detected by a receiving device and location data that indicates a location of the receiving device;
providing, by the computing device, the media data that represents the item of media content and the location data that indicates the location of the receiving device as an input to a model that is configured to determine whether the item of media content likely includes deepfake content;
receiving, by the computing device and from the model, data indicating whether the item of media content likely includes deepfake content; and
based on the data indicating whether the item of media content likely includes deepfake content, determining, by the computing device, whether the item of media content likely includes deepfake content.

2. The method of claim 1 , comprising:
receiving, by the computing device, biometric data that reflects an attribute of an additional user while a receiving device detected the item of media content or while the receiving device outputted the media data that represents the item of media content,
wherein determining whether the item of media content likely includes deepfake content is further based on the biometric data that reflects the attribute of the additional user while the receiving device detected the item of media content or while the receiving device outputted the media data that represents the item of media content.

3. The method of claim 1 , comprising:
receiving, by the computing device, sensor data that reflects an attribute of a receiving device while the receiving device detected the item of media content or while the receiving device outputted the media data that represents the item of media content,
wherein determining whether the item of media content likely includes deepfake content is further based on the sensor data that reflects the attribute of the receiving device while the receiving device detected the item of media content or while the receiving device outputted the media data that represents the item of media content.

4. The method of claim 1 , wherein the model is trained using machine learning and training data that includes a plurality of items of media content that are each labeled as including deepfake content and corresponding location data that indicates a location of each receiving device that detected each item of media content of the plurality of items of media content.

5. The method of claim 1 , wherein:
receiving the data indicating whether the item of media content likely includes deepfake content comprises:
receiving data indicating that the item of media content likely includes audio of the user's voice that was not spoken by the user, and

determining whether the item of media content likely includes deepfake content comprises:
determining that the item of media content likely includes deepfake content based on the data indicating that the item of media content likely includes audio of the user's voice that was not spoken by the user.

6. The method of claim 1 , wherein:
receiving the data indicating whether the item of media content likely includes deepfake content comprises:
receiving data indicating that the item of media content likely includes video of the user that depicts actions of the user that were not performed by the user, and

determining whether the item of media content likely includes deepfake content comprises:
determining that the item of media content likely includes deepfake content based on the data indicating that the item of media content likely includes video of the user that depicts actions of the user that were not performed by the user.

7. The method of claim 1 , wherein:
receiving the data indicating whether the item of media content likely includes deepfake content comprises:
receiving data indicating that the item of media content does not include audio of the user's voice that was not spoken by the user and does not include video of the user that depicts actions of the user that were not performed by the user, and

determining whether the item of media content likely includes deepfake content comprises:
determining that the item of media content likely does not include deepfake content based on the data indicating that the item of media content does not include audio of the user's voice that was not spoken by the user and does not include video of the user that depicts actions of the user that were not performed by the user.

8. The method of claim 1 , comprising:
receiving, by the computing device, additional media data that represents the item of media content;
providing, by the computing device, the additional media data that represents the item of media content as an additional input to the model; and
receiving, by the computing device and from the model, additional data indicating whether the item of media content likely includes deepfake content,
wherein determining whether the item of media content likely includes deepfake content is further based on the additional data indicating whether the item of media content likely includes deepfake content.

9. The method of claim 1 , comprising:
receiving, by the computing device, data confirming whether the item of media content includes deepfake content; and
updating, by the computing device, the model using machine learning and using the data confirming whether the item of media content includes deepfake content and the item of media content and the location data that indicates the location of the receiving device.

10. A system, comprising:
one or more processors; and
memory including a plurality of computer-executable components that are executable by the one or more processors to perform a plurality of actions, the plurality of actions comprising:
receiving, by a computing device, media data that represents an item of media content detected by a receiving device and location data that indicates a location of the receiving device;
providing, by the computing device, the media data that represents the item of media content and the location data that indicates the location of the receiving device as an input to a model that is configured to determine whether the item of media content likely includes deepfake content;
receiving, by the computing device and from the model, data indicating whether the item of media content likely includes deepfake content; and
based on the data indicating whether the item of media content likely includes deepfake content, determining, by the computing device, whether the item of media content likely includes deepfake content.

11. The system of claim 10 , wherein the actions comprise:
receiving, by the computing device, biometric data that reflects an attribute of an additional user while a receiving device detected the item of media content or while the receiving device outputted the media data that represents the item of media content,
wherein determining whether the item of media content likely includes deepfake content is further based on the biometric data that reflects the attribute of the additional user while the receiving device detected the item of media content or while the receiving device outputted the media data that represents the item of media content.

12. The system of claim 10 , wherein the actions comprise:
receiving, by the computing device, sensor data that reflects an attribute of a receiving device while the receiving device detected the item of media content or while the receiving device outputted the media data that represents the item of media content,
wherein determining whether t

Assignees

• T Mobile Usa Inc

Inventors

• Yocam Eric
• Obaidi Ahmad Arash

Classifications

• G06N20/00

  Machine learning · CPC title

• G06N5/04 Primary

  Inference or reasoning models · CPC title

• G06F21/64

  Protecting data integrity, e.g. using checksums, certificates or signatures · CPC title

• G06F16/285

  Clustering or classification · CPC title

• G06N5/025

  Extracting rules from data · CPC title