ASPIRE: Accessible Slide Presentation via Intelligent Real-time Editing

Abstract

Live presentations remain difficult to follow for blind and low-vision (BLV) audiences: the speaker is audible, but slide titles, figures, and diagrams are not. Prior work introduced delay-buffered editing, which buffers a talk’s audio so redundant speech can be removed and concise slide descriptions inserted into the reclaimed gaps, but it was evaluated offline, with descriptions and edits made by hand. We present ASPIRE, a fully automated, real-time realization for live talks: it captures the presenter’s screen and audio, detects slide changes, generates and speaks a short description of each new slide, and mixes it in at a natural pause, continuously reclaiming the inserted time by trimming over-long pauses and filler words so the listener stays close to live, all streamed to listeners’ own devices with no special hardware. Across four full-length talks and a pilot with four BLV participants, content-recall accuracy rose from 16.7% to 91.7% with ASPIRE, alongside large gains in subjective slide understanding.