Portfolio
Real‑Time Sign Language to Speech Translation System: Low‑Latency Video‑to‑Voice Pipeline
An integrated accessibility pipeline that captures live video, extracts pose/hand/face keypoints, models gesture sequences, outputs text, and synthesizes speech for real-time interaction.
Project details
What we delivered
Overview
Real-time capture → keypoint extraction → temporal modeling → text decoding → speech output; designed for sub-second responsiveness with an offline-capable mode.
Why Keypoints (Not Raw Video)
Keypoints are lighter, faster, more stable for temporal modeling, and reduce privacy risk compared to storing/processing full frames.
Pipeline
Webcam/mobile input → buffer frames → MediaPipe Holistic / OpenPose → temporal windowing → LSTM/Transformer → decode → TTS → audio output (+ optional transcript).
Use Cases
Healthcare front desks and triage
Customer service and government counters
Education support
Workplace HR and onboarding