08 / Case study
SilentSpeaker
A browser-based, real-time translator between ASL signs and text or speech.
- TensorFlow.js01
- MediaPipe02
- FastAPI03
Calgary, ABMMVI KID
Same city.Brightertomorrows.
The challenge
How can sign-language translation run quickly and privately in an ordinary web browser?
Impact
- Reported model accuracy
- ~94%
- Labeled gestures collected
- 7,000+
Calgary
Realplacesrealprogress
Overview
SilentSpeaker was built to reduce communication barriers between ASL users and people who communicate through text or speech. It supports sign-to-text or speech and text-or-speech-to-sign modes.
Build
Our team manually collected and verified more than 7,000 labeled hand-gesture samples. The recognition pipeline uses MediaPipe and TensorFlow.js so the lightweight model can run in the browser, with FastAPI supporting the wider application.
Result
The exported project record reports approximately 94% model accuracy. The January 2026 prototype focused on responsive, dependency-light translation.