"VectorZero team on AMD Developer Hackathon: ACT III Hackathon"

Team Idea

CUE (Constrained Utterance Engine) is an AI-powered accessibility platform that automatically transforms video into timing-aware, multi-audience audio-described content. Unlike conventional AI narration, CUE treats time as a hard constraint: it understands each scene, identifies dialogue-free gaps, generates multiple narration candidates, synthesizes them, measures their actual duration, and selects the most informative description that fits without overlapping dialogue. The system combines speech recognition, vision-language models, language models, text-to-speech, grounding verification, and deterministic scheduling in a GPU-accelerated pipeline running on AMD infrastructure. A single video analysis can be reused to generate different experiences, such as standard audio description, simplified narration for children, and multilingual versions.