“Our aim is a digital world where people can believe what they see and hear”

SIDN Fund supports development of DuckDuckGoose's deepfake detection tools

Digitized representation of a face

Deepfake technology is developing rapidly. Computers are now able to synthesise photos, videos, audio clips and even voices that are almost impossible to distinguish from the real thing. Synthetic material can be found in the media, and in the context of digital identity verification. What's more, deepfakes are amongst the most influential sources of disinformation. DuckDuckGoose therefore develops tools for identifying deepfakes and distinguishing between real and synthesised material. One of the organisation's most recent projects involved development of a tool for detecting synthetic sound fragments. "Our mission is to help build a digital world where people can believe what they see and hear," says Joris Mollinga, co-founder of DuckDuckGoose. Joris talks to us about his organisation's contribution to controlling disinformation and understanding the deepfake phenomenon.

Synthetic speech detection

In recent years, huge advances have been made in the field of sound synthesis. "Along with photo and video, sound plays an increasingly important role in the world of deepfakes," explains Joris. "With real-time voice cloning technology, it's possible to simulate a person's voice on the basis of just a few seconds of audio samples. Artificial intelligence (AI) can then be used to replicate the sound of an individual saying anything. And malicious use of such voice cloning technology is on the rise. It's already common for internet users to get WhatsApp messages from scammers posing as relatives in need of financial assistance. Most people who get a written message are going to sense that something isn't right, of course. But what if the message is followed up by a cloned voice call repeating the request? It's easy to imagine people being taken in."

Joris Mollinga, Mark Evenblij and Parya Lotfi, founders of DuckDuckGoose

Figure 1: Joris Mollinga (left) with co-founders Mark Evenblij (middle) and Parya Lotfi (right).

"SIDN Fund has provided financial support enabling us to develop a tool for detecting synthetic sound fragments to complement our existing detection products. The project is just getting started, and is scheduled to run until April 2024. Our ultimate goal is to make a tool available that investigators, journalists and internet users can use themselves to detect synthetic sound fragments."

Elise van Schaik, Project Coordinator at SIDN Fund: "DuckDuckGoose is helping to restore public confidence in information on the internet. The deepfake detection tool and the new tool they're developing for detecting synthetic sound fragments enable professionals and other internet users to automatically distinguish fakes from the real thing. That kind of capability is very important in a world blighted by growing disinformation and misinformation problems. The detection tools offered by DuckDuckGoose are helping to make internet users more aware of deepfakes and the fact that not everything on the internet can be trusted."

Previous support for DuckDuckGoose from SIDN Fund

"SIDN Fund previously supported our Pioneer Project involving the detection of deepfake videos," continues Joris. "The latest grant enables us to develop the concept further. Internet users can now use our products and services at reduced rates, or even for free. Once we had developed our deepfake detection tool, SIDN Fund helped us with the follow-up initiative, DeepfakeProof: a free browser plugin for the automatic detection of deepfakes. With the plugin installed, the user gets a notification if they look at a deepfake photo or video in their browser."

As well as providing financial support for product development, SIDN Fund has helped DuckDuckGoose forge valuable alliances. "Through SIDN Fund's network, we were introduced to a lawyer who specialises in privacy law," says Joris. "That was really helpful, because we were getting lots of questions about the privacy aspects of our browser plugin. The lawyer helped us to answer those questions, and has since become our legal counsel."

The future of deepfakes

"Voice cloning technology is getting more sophisticated all the time. We're also seeing the rise of text-to-image software that can convert written words into realistic pictures. Text-to-video capability is likely to follow. The user will be able to type 'car of make [x] with registration number [y] hits tree', and the software will generate a realistic video of the event described. Developments in this field are moving incredibly quickly. I expect that, within a few years, you'll be able to do things like that on your phone. So there's an ongoing need for detection product development. Using our software, an organisation can check the authenticity of material they receive and spot any deepfakes. That's an important breakthrough. Our AI solutions for the detection of deepfakes in audiovisual content recently won the Identity Innovation Award." The award is presented to the developers of a solution that will shape the future of digital identities.

"It's important that internet users are more aware of deepfakes, and realise that not every photo, video or audio clip is what it seems to be. Widespread use of our products can help to build that awareness," concludes Joris.

For more about DuckDuckGoose and its detection products, visit www.duckduckgoose.ai or mail Joris Mollinga at joris@duckduckgoose.nl.