Deepfake detection models utilizing subject-specific libraries

Patent metadata

Field	Value
Publication number	US-12475695-B2
Application number	US-202117481475-A
Country	US
Kind code	B2
Filing date	Sep 22, 2021
Priority date	Sep 22, 2021
Publication date	Nov 18, 2025
Grant date	Nov 18, 2025

Abstract

An apparatus to facilitate deepfake detection models utilizing subject-specific libraries is disclosed. The apparatus includes one or more processors to store a plurality of deepfake detection models corresponding to a plurality of subjects of interest; receive a query to identify whether data pertaining to a target subject of interest is a deepfake, the target subject of interest comprised in the plurality of subjects of interest and associated with a subject identifier (ID); identify a deepfake detection model corresponding to the subject ID; extract features for deepfake detection from the data; input the extracted features to the identified deepfake detection model corresponding to the subject ID; and responsive to an output of the deepfake detection model exceeding a determined deepfake threshold, generate a notification, in response to the query, indicating a possible deepfake attack corresponding to the target subject of interest.

First claim

What is claimed is:

1 . An apparatus comprising:
one or more processors to:
store a plurality of deepfake detection models corresponding to a plurality of subjects of interest, wherein the plurality of deepfake detection models are each generated using subject-specific data collected for a corresponding subject of interest of the plurality of subjects of interest and the plurality of deepfake detection models are trained using the subject-specific data to distinguish synthetic data from authentic data corresponding to the plurality of subjects of interest;
receive a query to identify whether data pertaining to a target subject of interest is a deepfake, the target subject of interest comprised in the plurality of subjects of interest and associated with a subject identifier (ID);
identify, using the subject ID, a deepfake detection model corresponding to the subject ID from the plurality of deepfake detection models;
extract features for deepfake detection from the data pertaining to the target subject of interest, wherein the features comprise a combination of face, voice, and text corresponding to the target subject of interest;
input the features to the deepfake detection model corresponding to the subject ID; and
responsive to an output of the deepfake detection model satisfying a determined deepfake threshold, generate a notification, in response to the query, indicating a possible deepfake attack corresponding to the target subject of interest;
wherein the plurality of deepfake detection models are stored on a local computing device and trained at the local computing device based on training features extracted from communication data with the plurality of subjects of interest at the local computing device and based on synthetic deepfake data of the plurality of subjects of interest created at the local computing device.

2 . The apparatus of claim 1 , wherein the plurality of subjects of interest comprise people.

3 . The apparatus of claim 1 , wherein the data comprises at least one of images, video, audio, text, activity patterns, or mannerisms.

4 . The apparatus of claim 1 , wherein a deepfake detector on the local computing device is to interface with one or more communication applications executing on the local computing device to obtain the communication data to train the plurality of deepfake detection models.

5 . The apparatus of claim 4 , wherein the plurality of deepfake detection models are transmitted to a public computing device, the plurality of deepfake detection models at the public computing device utilized for access by a plurality of users to verify authenticity of media submitted by the plurality of users.

6 . The apparatus of claim 5 , wherein the plurality of deepfake detection models are trained offline from the public computing device.

7 . The apparatus of claim 1 , wherein the plurality of deepfake detection models comprise classifiers to distinguish the synthetic data from the authentic data.

8 . The apparatus of claim 1 , wherein the plurality of deepfake detection models comprise at least one of distribution models, classifiers, or generative models.

9 . A non-transitory computer-readable storage medium having stored thereon executable computer program instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:
storing a plurality of deepfake detection models corresponding to a plurality of subjects of interest, wherein the plurality of deepfake detection models are each generated using subject-specific data collected for a corresponding subject of interest of the plurality of subjects of interest and the plurality of deepfake detection models are trained using the subject-specific data to distinguish synthetic data from authentic data corresponding to the plurality of subjects of interest; receiving a query to identify whether data pertaining to a target subject of interest is a deepfake, the target subject of interest comprised in the plurality of subjects of interest and associated with a subject identifier (ID); identifying, using the subject ID, a deepfake detection model corresponding to the subject ID from the plurality of deepfake detection models; extracting features for deepfake detection from the data pertaining to the target subject of interest, wherein the features comprise a combination of face, voice, and text corresponding to the target subject of interest; inputting the features to the deepfake detection model corresponding to the subject ID; and responsive to an output of the deepfake detection model satisfying a determined deepfake threshold, generating a notification, in response to the query, indicating a possible deepfake attack corresponding to the target subject of interest; wherein the plurality of deepfake detection models are stored on a local computing device and trained at the local computing device based on training features extracted from communication data with the plurality of subjects of interest at the local computing device and based on synthetic deepfake data of the plurality of subjects of interest created at the local computing device.

10 . The non-transitory computer-readable storage medium of claim 9 , wherein the plurality of subjects of interest comprise people.

11 . The non-transitory computer-readable storage medium of claim 9 , wherein the data comprises at least one of images, video, audio, text, activity patterns, or mannerisms.

12 . The non-transitory computer-readable storage medium of claim 9 , wherein a deepfake detector on the local computing device is to interface with one or more communication applications executing on the local computing device to obtain the communication data to train the plurality of deepfake detection models.

13 . The non-transitory computer-readable storage medium of claim 12 , wherein the plurality of deepfake detection models are transmitted to a public computing device, the plurality of deepfake detection models at the public computing device utilized for access by a plurality of users to verify authenticity of media submitted by the plurality of users.

14 . The non-transitory computer-readable storage medium of claim 9 , wherein the plurality of deepfake detection models comprise classifiers to distinguish the synthetic data from the authentic data.

15 . The non-transitory computer-readable storage medium of claim 9 , wherein the plurality of deepfake detection models comprise at least one of distribution models, classifiers, or generative models.

16 . A method comprising:
storing, by a processing device, a plurality of deepfake detection models corresponding to a plurality of subjects of interest, wherein the plurality of deepfake detection models are each generated using subject-specific data collected for a corresponding subject of interest of the plurality of subjects of interest and the plurality of deepfake detection models are trained using the subject-specific data to distinguish synthetic data from authentic data corresponding to the plurality of subjects of interest; receiving a query to identify whether data pertaining to a target subject of interest is a deepfake, the target subject of interest comprised in the plurality of subjects of interest and associated with a subject identifier (ID); identifying, using the subject ID, a deepfake detection model corresponding to the subject ID from the plurality of deepfake detection models; extracting features for deepfake detection from the data pertaining to the target subject of interest, wherein the features comprise a combi

Assignees

• Intel Corp

Inventors

• Stemmer Georg
• Marshall Carl
• Srivastava Satyam
• Demir Ilke

Classifications

• G06N20/00

  Machine learning · CPC title

• G06V20/41

  Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items (segmenting video sequences G06V20/49) · CPC title

• G06V20/30

  in albums, collections or shared content, e.g. social network photos or video · CPC title

• G06F18/2148

  characterised by the process organisation or structure, e.g. boosting cascade · CPC title

• G06F18/285

  Selection of pattern recognition techniques, e.g. of classifiers in a multi-classifier system · CPC title