ZSL Parliament Interpreter: Design and Implementation of a Speech-to-Sign Language Interpretation System for the National Assembly of Zambia

Access to parliamentary debate requires communication channels that accommodate different hearing and language needs. This project presents a browser-based speech-to-sign interpretation prototype for the National Assembly of Zambia. The system captures spoken English, produces a streaming transcript, generates an intermediate gloss representation and plays available recorded movements through a three-dimensional avatar. The system combines Flask, Flask-SocketIO, SQLAlchemy, NVIDIA NeMo streaming automatic speech recognition, a dictionary-based gloss engine and Three.js/VRM avatar rendering. The application contains 1916 word mappings, 25 phrase mappings and 1941 recorded motion sequences across 28 categories, including alphabet animations for fingerspelling, alphabet word from a to a and a number category. Authentication, email verification, recording permissions, profile management and administrative controls support operation within one application. Verification established selected software behaviours through component and browser checks, JavaScript regression scripts, and server-side checks of speech processing and asset delivery. Additional formative evaluations examined speech transcription and avatar playback. These results support component execution and integration, but do not establish reliable Zambian Sign Language (ZSL) translation or parliamentary readiness. The project is an engineering prototype; language-specific validation and comprehension studies with intended users remain priorities.