With the advancement of digital recording devices, there is an emerging trend of citizen journalism/ collaborative media where the public shares or contributes information directly to the mainstream media channels. Coupled with the
emergence of AI video generators, even amateurs and non-coders can produce realistic avatars and videos to spread fake news.
Mainstream media needs to consider their professional and ethnical obligations when reporting or sharing this contribution on their channels thus a solution is needed to decipher the authenticity of such information.
The Sleuth is I²R's latest technology in deciphering such edited video and even text. It can identify the exact segment and localise the video or text manipulation. This provides an additional layer of verification and
protection against fake news.
Unlike other technologies, we have developed three different detection engines for three different modalities (Audio, Video, LipSync) of manipulations and achieved state-of-the-art accuracies. Our face manipulation-based deepfake detection
technology achieved minimum accuracy of 90% on a blind dataset.
Example of The Sleuth detecting alterations in a video
Key Features of The Sleuth includes:
•
Detects alteration in video, voice and text automatically.
•
Verifies contents (video, image and text) with the original online.
•
Highlights the manipulated content accurately even if it is edited partially.
Here's a short video about The Sleuth.
WhatsApp and Telegram is one of the most common platforms that scammers use to contact victims in scams according to the Singapore Police Force Scams and Cybercrime Statistics 2023.
To tackle this rising issue, The Sleuth is now also accessible on WhatsApp and Telegram to empower users to spot fake news or scam messages.
Users will be able to send texts to The Sleuth platforms to verify it is factually true or false, and detects any hateful speech or scam information. The content will be analysed and replied with additional details to support the results. Depending on the length of text provided by the users, results are generated in matter of seconds.
Works are in progress to advance this onto desktop browser extensions.
