Audio-Video Deepfake Detection Using Explainability Feedback

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Deepfakes have become increasingly realistic due to advancements in artificial intelligence, making it difficult for humans to accurately distinguish between authentic and manipulated media, posing risks such as fake news and fraud.

Innovation Solution

Utilizing AI models to analyze audio and video features of media files, comparing sound classifications from both modalities to determine similarity within a threshold, and implementing a neural network-based system to classify media as authentic or deepfake, with feedback mechanisms for model tuning.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Illumination intensity

If AI models are used to generate deepfake media, then media realism and visual quality are improved, but media authenticity and detectability deteriorate

Engineering Contradiction:
Improvemedia realismVSAvoidmedia authenticity
Core Design Contradiction:
Illumination intensityVSReliability

Solution Approach 1:

The patent introduces an intermediary detection system that analyzes audio-visual synchronization patterns to identify deepfakes. This mediator compares expected synchronization (from authentic media) with actual synchronization (from deepfake media) to detect anomalies, thereby restoring detectability without affecting the generation quality of deepfakes.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent implements feedback mechanisms where detection results are used to refine and improve the detection models. By continuously analyzing audio-visual discrepancies and feeding this information back into the system, the detection accuracy improves over time while maintaining the ability to distinguish authentic from deepfake media.

Inventive Principle:
Principle #23Feedback

2Measurement precision

If audio and video analysis are combined for deepfake detection, then detection accuracy is improved, but system complexity increases

Engineering Contradiction:
Improvedetection accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent merges audio and video analysis into a unified detection framework that evaluates audio-visual synchronization. By combining these modalities and analyzing their temporal relationships, the system achieves higher detection accuracy than single-modality approaches while managing complexity through integrated processing.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent segments the detection process into distinct components: audio feature extraction, video feature extraction, synchronization analysis, and anomaly detection. This segmentation allows each component to be optimized independently while maintaining overall system accuracy and making the complex system more manageable and interpretable.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP4050571B1Methods and apparatus to perform deepfake detection using audio and video features
Publication Date: 2026.02.18 MCAFEE LLC
  • EP4050571B1 patent drawingFigure 1
  • EP4050571B1 patent drawingFigure 2
  • EP4050571B1 patent drawingFigure 3A~3B

AI summary

Methods, apparatus, systems and articles of manufacture to improve deepfake detection with explainability are disclosed. An example apparatus includes a deepfake classification model trainer to train a classification model based on a first portion of a dataset of media with known classification information, the classification model to output a classification for input media from a second portion of the dataset of media with known classification information; an explainability map generator to generate an explainability map based on the output of the classification model; a classification analyzer to compare the classification of the input media from the classification model with a known classification of the input media to determine if a misclassification occurred; and a model modifier to, when the misclassification occurred, modify the classification model based on the explainability map.