Rep. Ro Khanna has named Kushaagra Gupta from Irvington High School as the winner of the 2021 Congressional App Challenge in California’s 17th District.
When asked what inspired the creation of Quick Dub, the student said, “I am an avid competitive programmer, so I enjoy learning about various types of algorithms. While scrolling through Youtube, I noticed a large playlist of videos on competitive programming algorithms. I clicked on the video and the comment section was flooded with great reviews about how good the lecturer was. The lecturer started to speak and to my dismay he was speaking in fluent Russian. That’s when I realized a great deal of content, especially educational content, has been locked behind this language barrier.
So, I set out to create a dubbing application that would allow myself and others to break through this barrier. I used various translation APIs from google and created the first version of Quick Dub. It was able to translate any video, breaking the language barrier but something was still off: the dubbed content just wasn’t as natural as the real version. The speaker’s facial expressions and lip movement didn’t match what they were saying.
For a long time I was stumped. How could I make these dubs look more natural? I researched various methods of lip-syncing, and I found less robotic voices for the Google Text-to-Speech API, but it still wasn’t perfect. Soon, I turned my focus towards other things, and left the project in the depths of my hard drive. I started learning about Artificial Intelligence, Machine Learning, and Neural Networks. One day while learning about Generative Adversarial Networks (GANs), I had a Eureka moment.
What if I used a GAN in Quick Dub?
I immediately got to researching and after a couple days I found Wav2Lip. Wav2Lip was a lip generation model that could manipulate the speaker’s face to follow inputted audio. After weeks of trial and error, I had integrated it into Quick Dub and got it working. Looking back, this was the most important feature in my app because it separates Quick Dub from all the other apps on the market.
All in all, the tremendous problem of eliminating the language barrier is what inspired me to create this app. I find this problem to be extremely exciting, and I hope we make more progress in the coming years!”
The 2021 Congressional App Challenge yielded 2,101 fully functioning apps. After eighteen months of disruptions to educational cadences for students everywhere, the Congressional App Challenge came roaring back with 7,174 students registering for this year’s competition. All told, 340 Members of Congress hosted Congressional App Challenges in their districts across 50 states, Puerto Rico, Guam, the Mariana Islands, the U.S. Virgin Islands, and Washington, D.C.
The Congressional App Challenge is an official initiative of the U.S. House of Representatives, where Members of Congress host contests in their districts for middle school and high school students, encouraging them to learn to code and inspiring them to pursue careers in computer science. Each participating Member of Congress selects a winning app from their district, and each winning team is invited to showcase their winning app to Congress during our annual #HouseOfCode festival. The program is a public-private partnership made possible through funding from Omidyar Network, AWS, theCoderSchool, Facebook, Replit, Accenture, and others.
The 2022 Congressional App Challenge will launch in June of 2022, and eligible students can pre-register for the competition now.