NEXT-GEN LOCAL AI • 100% IN-BROWSER

Bridge Sign Language & Speech Instantly & Locally

SignSpeak transforms camera gesture inputs into natural spoken audio in real time. Powered by MediaPipe 3D landmark tracking, TensorFlow.js, IndexedDB, and Web Speech API.

Start Live Demo Record Local Gestures

Zero Server / API Calls

MediaPipe & TFJS Accelerated

Native Voice Output

Application Modules

High-performance AI tools running natively in your browser workspace.

Centerpiece Demo

Live Inference Engine

Real-time sign language recognition with sliding-window majority voting, confidence gating, and Web Speech sentence voice output.

Zero Server Required

Local Dataset Collector

Capture static gesture bursts or continuous dynamic sign sequences. Normalized wrist-relative data stored directly in IndexedDB.

21 Joint Skeleton

MediaPipe Landmark Tracker

Inspect 21 3D hand landmarks in real time with high-fps skeleton overlays and raw coordinate debugging panels.

How SignSpeak Works

Three seamless steps from camera frames to natural speech.

Track Hand Skeleton

MediaPipe Hands tracks 21 3D joint landmarks live in browser via webcam.

Normalize & Store

Landmarks are scale-normalized and saved to IndexedDB or trained TFJS models.

Translate & Speak

Signs are reordered into natural grammar and spoken aloud via Web Speech TTS.