Inspiration
I was scrolling on tiktok and I heard a lot about scam calls and how people are getting fooled thinking it's their family members due to technological advancements in Deep Fakes. Which lead me to creating Faking the Deep.
What it does
Faking the Deep analysis videos, audios, and body movements and tells the users if it's Generative AI/ Deep Fakes or not. It doesn't just say yes or no, it gives how confident it is in it's answer so that there is a probality.
Challenges we ran into
When I was using Python 3.14 it was fairly new so the TensorFlow and Open CV were very confused with it and my code kept crashing. Which was really frustrating. Also I've never worked Arduino before so wiring the LCD screen was really hard and it took me so long to get the wiring right. When I was trying to get the audio code to work, when I inputted an audio file that had no talking it kept saying the silence meant the video was fake so I had to make a speech activity checker to filter that out. Gemini's ratings for a hot second was the totally opposite of what it meant so I had to quickly fix that as well. I think my major challenge was trying to get everything hosted and online. I tried using Google Cloud Run, Huggies.co but it didn't seem like any of them liked the big file.
Accomplishments that we're proud of
I've never done such a large project and surly not by myself so I am glad that I was able to make this multi-layer deepfake detector. I am also proud that I didn't just let the code sit on a web browser but I took into the physical world a little bit with the LCD display. I thought that was a fun addition. What I am most proud of though is that I managed to debug all the files, codes, etc. And that I didn't give up, even when I wanted to.
What we learned
I learned how to manage python virtual environments, enhanced my ability to read error messages. I also learned that pre trained models can be combined into a single pipeline. Which might seem basic but I am beginner so. Most free hosting comes some costly tradeoffs weather it's less CPU, storage, etc.
What's next for Faking the Deep
I really want to be able to implement Faking the Deep into real time video calls, phone calls, it seems hard to tap into phone calls. But with connections it's possible. I also want to improve the face detection and overall detection. In addition make the LCD aurdino bluetooth.