SilentLink

Since you asked for SilentLink (the Accessibility project), here is a complete, Devpost-ready draft. It uses the same catchy, professional, yet "hacker-style" tone as the LipsLips Revolution example you provided.

Inspiration

Communication is a human right, but for millions with speech impairments (like ALS, vocal cord damage, or stroke survivors), it’s a daily struggle. We noticed that while Voice Assistants are everywhere, they are useless if you can't speak out loud. We wanted to build a bridge—a way to turn silent intent into audible reality. We took the concept of "lip-reading" and turned it into a life-changing tool.

What it does

SilentLink is a real-time "Digital Voice" for those who cannot speak. It allows a user to communicate naturally just by moving their lips—no sound required.

The Silent Mic: The app uses your camera to "hear" what your lips are saying.

Instant Translation: It converts lip movements into text with high accuracy using the Symphonic Labs API.

The Personal Voice: Instead of a robotic voice, it uses ElevenLabs to speak the words in a natural, human-like tone (or even the user's own cloned voice).

Speed to Speech: It’s designed for real conversations, focusing on low latency so there’s no awkward silence.

How we built it

SilentLink is a high-performance web app built with Next.js.

Lip-to-Text: We integrated the Symphonic Labs API to process video frames and extract spoken words.

The "Voice": We used ElevenLabs API for the text-to-speech engine to ensure the output sounded like a person, not a machine.

Real-time Engine: To avoid the "upload-and-wait" delay, we built a Node.js backend using WebSockets to stream video data chunks for faster processing.

Database: MongoDB stores user profiles and "Quick Phrases" for common daily needs (like "I need water" or "Thank you").

Challenges we ran into

The biggest hurdle was Latency. Initially, sending a 10-second video and waiting for a response took nearly 40 seconds—far too slow for a real conversation. We had to dive deep into Network Inspection to see how modern video streaming works. By switching from standard POST requests to a Streamed WebSocket approach and optimizing our video encoding (reducing resolution without losing lip clarity), we cut the delay by 75%.

Accomplishments that we're proud of

We managed to create a working prototype that can recognize complex sentences in under 5 seconds. But more than the tech, we are proud of creating something that has the potential to give someone their "voice" back. Seeing the AI perfectly speak a sentence that was only "mouthed" silently felt like magic.

What we learned

We learned a massive amount about Computer Vision and the limitations of browser-based video processing. We also sharpened our skills in API Reverse Engineering and learned how to optimize cloud-based AI workflows to feel "instant" for the end-user. Most importantly, we learned that technology is at its best when it solves a real human problem.