Digger will provide functionalities in Truly Media to detect audio tampering in videos.
What you are facing here is a deepfake: machine-learning technology used to manipulate or fabricate audio and video to make people do and say things that they never did or said. They can be extremely realistic and convincing but are actually fake.
It is challenging to verify content which has been stripped of context, edited, mislabeled or staged. What is even more complex is to verify content that has been modified. Synthetic audiovisual media can be generated with a variety of deep learning techniques. Although the technology is still improving and not sophisticated it is still widely available.
In the Digger project we will develop software together with Deutsche Welle, ATC and Fraunhofer IDMT to detect deepfakes by using state-of-the-art audio forensics technologies.
Our approach makes use of:
- Microphone analysis
- Electrical Network Frequency analysis
- Codec analysis
- Audio segment matching
These functionalities will be integrated in Truly Media and made available to verification experts via API.
In the meantime: The Digger project also aims to develop a community to share knowledge and initiate collaboration in the field of deepfake detection. Want to join? Follow us on Twitter: @digger_project. See you there!