Logo Lanfrica

k-atror/GSL-Translator

Domain:

natural language processing

Record type:

project
Creator:
k-a
Host:
An AI project aimed at reducing communication barriers faced by deaf and hard-of-hearing communities in Ghana, with a focus on healthcare accessibility. # Ghana Sign Language (GSL) Translator 🚧 This repository contains an ongoing project developing a real-time Ghana Sign Language (GSL) translator aimed at improving communication between deaf or hard-of-hearing individuals and hearing communities in Ghana, particularly in healthcare and public accessibility contexts. ## Project Status 🚧 **Work in Progress** The project is actively under development. The focus is on experimentation, system design, and iterative improvement rather than delivering a finished product. ## System Overview The GSL Translator is designed as a multi-stage AI pipeline: Camera Input → Exported Teachable Machine Model → Word-Level Predictions (in app) → Timing & Buffering Logic → Sentence Construction Pipeline → Text / Speech Output ### 1. Sign Recognition - Uses **Google Teachable Machine** to train a computer vision model that classifies individual GSL hand signs. - Sign selection is informed by the Ayele Foundation GSL Dictionary. - The trained model is exported and used in the application to produce **word-level predictions** in real time. ### 2. Timing and Buffering Logic - Ensures that detected signs are **held long enough** to count as a word. - Prevents repeated detections of the same sign. - Detects pauses to signal sentence boundaries. ### 3. Sentence Construction - Accumulated word outputs are processed through an AI pipeline. - Combines rule-based logic and language-model techniques to **approximate GSL grammatical patterns** and convert sequences into clear English sentences. ## Tools and Technologies - Google Teachable Machine (computer vision training) - Python / JavaScript for pipeline logic and app - Real-time webcam input - Language models for sentence construction (experimental) ## Limitations - Currently supports only **basic signs and short phrases**. - Facial expressions and non-manual markers are not modeled. - Full linguistic translation of GSL grammar is outside the current scope. ## Future Work - Expand the sign …