SUSI Interpretation Project

This project aims to bring real-time, AI-powered live translation to the eventyay-video platform. Currently, the SUSI Translator prototype works well but relies on local computer microphones and standalone scripts. To make it ready for live online events, this project will upgrade it into a fast, web-native system that works entirely in the browser. The solution captures live audio directly from the speaker's browser, immediately filtering out silence to save server costs. This speech is then streamed at high speeds to a backend server. There, AI models (Whisper and an LLM) transcribe and translate the speech. To prevent robotic or broken grammar, the AI uses a "sliding window" to remember the last few spoken sentences, ensuring translations make sense in context. Finally, the translated text is sent back to the audience's screen as smooth, easy-to-read subtitles. Key Deliverables: - A browser-based audio capture and silence-filtering module. - A high-speed server connection (WebSockets) for instant data transfer. - A context-aware AI translation pipeline that prevents grammatical errors. - A stable, flicker-free subtitle interface built directly into the eventyay-video player.