SpeakEz: Giving Your Hands a Voice

Inspiration

“Imagine a world where every hand gesture finds its voice.”
Our goal was to bridge silence and sound, empowering every signer to be heard, whether ordering coffee or joining a conversation.

What it does

SpeakEz is a lightweight glove that reads finger-bend data and instantly speaks your ASL signs, no cameras or bulky apps required. It turns each gesture into clear, spoken words so communication flows naturally.

How we built it

  1. Browser-Based Classifier: Trained a CNN on hand-shape images for initial ASL letter recognition. This is used as a teaching tool for those interested in learning ASL.
  2. Flex-Sensor Glove Prototype: Integrated bend sensors into a glove to capture live finger positions.
  3. ML Model for Glove Data: Developed a time-series classifier to map sensor streams to ASL letters and words.
  4. Real-Time Translation: Streamed sensor data over Wifi, classified on the fly, and synthesized speech.

Challenges we ran into

  • Calibrating sensor drift across different hand sizes
  • Reducing classification latency
  • Building a robust training dataset for dynamic gestures

Accomplishments that we're proud of

  • Achieved 95% accuracy on flex-sensor letter classification
  • Demoed “I ❤️ YOU” and letter translation in under 2 seconds
  • Created an intuitive web quiz

What we learned

  • User-centric design is critical: simplicity beat complexity every time
  • Sensor fusion and data augmentation improve robustness
  • Real-time ML demands careful optimization of both model and hardware

What's next for SpeakEz

  • Expand vocabulary to include whole words and phrases
  • Add multilingual support for international sign systems
  • Refine glove ergonomics and battery life for all-day wear
Share this project:

Updates