 Thank you. This is joint work with my colleagues Daniel Genkin, Mihir Patami and Roy Schuster and it addresses a problem of grave importance for many researchers, which is the following. When we have that video conference with a colleague and they appear somewhat distracted or disengaged, is that because they are pondering the open problems that we have posed to them or because they are browsing the blockchain? Now, they probably have a webcam for the video conference, but it's pointing away from the screen. They also have a microphone placed in proximity to the screen, which raises the question, can we listen in on their screen? That is, are there acoustic side channels emanated by screens? To investigate this, we started by setting up a simple cover channel detection methodology. We rendered simple zebra stripes across the screen of various thickness and we recorded them using a microphone. And we observed that as we changed the thickness of the zebra stripes, the acoustic signature changes and the spectral line shift. Clearly, if different zebras have different acoustic signatures, there is image dependent leakage. You can observe this leakage on LED and LCD screens for many manufacturers and of different models, including old ones as well as fresh ones of Amazon. And in all of these screens, we can observe this acoustic van Eck, if you will, that reveals the content of the screens by the movements of the frequency contents. You can record this by many commodity microphones such as webcams or the entailer microphones built into many screens, phones, and to jump on the Internet of Things bandwagon, even your favorite digital assistant smart speaker that would also conveniently store that captured all of your own in the cloud for future reference. You can even measure these signals from afar, say, 10 meters away using our favorite parabolic dish and get the same signals. Now, what are these signals? To dig in, it's most convenient to start by a synthetic image. For example, this one portraying zebra stripes punctuated by a black rectangle. And if you record using a microphone, a screen as it's rendering this, you'll probably see an image similar to the following where that punctuation in the middle is clearly visible within the time trace during a single rendering of a single frame. And what we're actually observing for this screen is the line intensity modulated onto a 32 kilohertz acoustic carrier. In order to exploit this, you need to cope with additional challenges. There is significant frequency drift and time jitter, there is abundant noise, and there is redundancy between frames that we would like to exploit. We cope with this using custom digital signal processing algorithms and trace merging, piped into deep learning, using a convolutional neural networks, culminating in a classifier that can tell what your colleague is up to. The optimistic. We can this way distinguish not just the nature of the activity, but also which website is being browsed between 100 websites with excellent accuracy. We can extract text that is rendered on screen, and in case the victim is typing on a virtual on-screen keyboard, we can extract the keystrokes. All of this goes to show that indeed the picture is worth a thousand words, because by the time you've spoken a thousand words, we probably detected which picture you're looking at. And for additional words, please see our paper. Thank you.