Deepfakes Intensify The Proliferation of Misinformation
From Tom Hanks to MrBeast, recent reports of deepfakes among prominent public figures have reached an alarming rate. It seems like there isn’t a day where a deepfake related incident is not reported. The manipulated content isn’t just impacting entertainers, it’s seeped its way into politics and journalism. With the growing concern around the rise of misinformation, in this article, we explore deepfake technology and the top 3 deepfake detection tools that have been developed to help identify real from fake content.
What Are Deepfakes?
Deepfakes leverage AI techniques like machine learning and neural networks to fabricate audio, image, or video that falsely depict people saying or doing things they have never done. During the machine learning process (ML), the AI model trains on hours of authentic audio data of the target person to learn their speech patterns, facial features, and facial animation. It then uses this data to generate new synthetic media that realistically impersonate the target person.
How Deepfake Voice Is Created Using Text to Speech
Now that we have a general understanding of machine learning as the foundation of deepfake creation, we can focus on deepfake voice generation. The accessibility to voice AI generators, sometimes referred to as AI voice changers, has given individuals access to voice cloning. Any user can scrape someone’s voice data from the internet and upload that data into a voice AI generator where they can clone the target person’s voice. Once the voice cloning is complete, the user is able to generate AI voice content through text to speech (TTS) or speech to speech (STS) conversion. Below is a diagram of text to speech synthesis describing how the input (text) and speech (output) process evolves.
The text to speech process that takes place when generating voice AI content.
Voice Deepfake Detection – Resemble Detect
While on the topic of deepfake voice, we’ll look at our first tool, Resemble Detect. The AI model is an advanced AI deepfake detection tool that can identify manipulated audio and AI voices. Powered by a deep neural network, it analyzes audio data to uncover subtle clues of fabrication that are imperceptible to humans.
The detector creates spectrogram-like time-frequency embeddings represented in the diagram below, providing a comprehensive profile of the audio signal over both time and frequency dimensions. Supplying the network with the necessary time and frequency information reveals artifacts and inconsistencies in things like cadence, emphasis, and pacing that characterize AI manipulated speech.
Finally, the audio data is then run through a Deep Learning Model which processes the time and frequency data from the waveform. The results are then evaluated by a Classifier. The Classifier outputs a probability ranging from 0 to 1, with 1 indicating that there is a high probability that the audio is fake. By benchmarking against Resemble’s database of real human voices, the detector can reliably flag deepfakes, fake voices, and other audio manipulated by generative models.
Resemble Detect’s deepfake detection model under the hood.
With over 98% accuracy, Resemble Detect serves as a robust safeguard against the growing threat of hard-to-spot deepfake audio and synthesized voices spreading disinformation. Watch the real time demo of Detect analyzing a 25-second long deepfake audio clip below.
Resemble Detect’s voice deepfake detection model at work.
Within seconds the real-time AI voice detector gave a resounding positive prediction of 100%. Represented below is the deep neural network’s analysis of the deepfake audio file. The AI model analyzes the audio data in 2-second increments represented on the x-axis. The y-axis determines the probability of deepfake voice with the bold red line at 1.00 or 100%.
Resemble Detect’s analysis of the demo video above.
Video Deepfake Detection: FakeCatcher by Intel
In similar efforts, to address the rise of deepfake technologies, Intel has introduced its robust solution, FakeCatcher. Employing cutting-edge AI, FakeCatcher operates in real-time to accurately identify deepfake videos. This is a critical advantage in a world where misinformation can go viral in seconds. Unlike conventional deepfake detection tools that may rely solely on visual cues, FakeCatcher uses an advanced algorithm to analyze both physiological and visual signs. Using a technique called Photoplethysmography, referred to as PPG or the changing color of blood as it circulates through the body. An infrared light measures the volumetric variations of blood circulation. The PPG signals are taken from the face of the subject and converted to PPG maps and a deep learning approach is applied to classify whether the video is fake. Eye gaze based detection is also implemented to give the tool a 96% accuracy rating.
Video Deepfake Detection with FakeCatcher by Intel
Video & Image Deepfake Detection: Microsoft Video Authenticator Tool
Moving onto another technology powerhouse, Microsoft’s Video Authenticator Tool adds their own safeguards to deepfake detection online. This deepfake detection tool employs AI algorithms to combat not only video manipulations but images as well. The AI tool analyzes a subject looking for minor imperfections that in some cases can be invisible to the human eye. It also analyzes subtle fading and grey scaling elements to detect signs of manipulation. Similar to Resemble Detect, Microsoft’s solution provides a percentage-based ‘confidence score,’ offering a quantifiable measure of a video or image’s authenticity. With the escalating sophistication of deepfake video maker tools, robust solutions will be needed to reliably detect deepfake video.
Microsoft Video Authenticator Tool finds subtle imperfections in the subject being analyzed.
The Path Forward: Prioritizing Deepfake Detection
As deepfake technology grows more sophisticated, the threat of viral disinformation intensifies. But with advanced detection tools like Resemble Detect, FakeCatcher by Intel, and Microsoft Video Authenticator, we can work to expose manipulated media and counteract misinformation. Moving forward, tech companies, governments, platforms, and citizens must make ethical AI development and deepfake protection a priority. With diligence and cooperation, we can harness AI for good while blunting its risks. Robust detection gives us hope for a future where truth and trust still matter.