Sign in front of your phone. SignBridge does the rest.
SignBridge is a free Android app that translates American Sign Language (ASL) in real time. Sign in front of your phone’s camera, and the app speaks the translation aloud. The other person can speak back, and you’ll see captions on screen.
The whole thing runs on your phone. There’s no cloud, no upload, no waiting. The app is free to download and free to use during beta.
The four steps
The camera sees your hands
Hold your phone up. The front-facing camera tracks your hand and body position 30 times per second.
The model recognizes the sign
An on-device neural network turns the motion into a label. No internet connection needed.
Your phone speaks for you
The label is read aloud. Loud enough to hear in a normal conversation, soft enough not to be jarring.
You see what they say
The other person speaks; the phone transcribes. Conversation flows in both directions.
Who SignBridge is for
Deaf and hard-of-hearing signers
Talk to anyone — in a coffee shop, at the doctor’s office, at a bus stop — without needing an interpreter for everyday conversations.
Hearing people who want to talk with someone deaf
Conversations with deaf family, friends, colleagues, and customers, without leaning on writing or guesswork.
What it doesn’t do
Honest limits matter as much as features. Here’s what SignBridge isn’t:
- Not a replacement for a human interpreter in legal, medical, or other high-stakes contexts. Bring a certified interpreter for those.
- Currently ASL only. BSL, JSL, and other sign languages are on the roadmap but not yet supported.
- Some signs work better than others. The model is still learning. Common everyday signs are accurate; rarer signs need more contributor data.
- Requires a phone with a camera and decent lighting. Bright sun behind you, very dark rooms, or extremely shaky hands degrade accuracy.
How it gets better
Recognition accuracy depends on how much real signing the model has seen. Contributors record themselves signing a target word, the app sends only the abstract hand-position numbers (no video, ever) to our server, and the model retrains on the combined data on a regular cadence. More signers across more backgrounds means better accuracy for everyone.
Contribute signs