Near-Real-Time Subtitle-to-Speech Translation Using Mobile Vision: A Lightweight Assistive Accessibility Prototype

This paper presents a proof-of-concept subtitle-to-speech accessibility system designed to support multilingual subtitle reading through lightweight mobile vision techniques. The prototype integrates optical character recognition (OCR), optional machine translation, and text-to-speech rendering within a near-real-time processing pipeline. Rather than proposing a novel OCR or translation architecture, the work explores the feasibility of combining existing lightweight technologies into a practical accessibility-oriented framework suitable for future assistive applications. The system operates by restricting processing to a predefined subtitle region, reducing computational overhead and improving OCR stability under controlled conditions. Preliminary tests demonstrate the operational viability of the prototype for subtitle detection, translation, and speech rendering, while also highlighting limitations related to latency, OCR reliability, subtitle quality, and multilingual robustness. The work contributes to ongoing research in lightweight assistive technologies and multimodal accessibility systems.