Voice-Enabled Wearable System for Visual Accessibility for the Visually Impaired

Abstract

Visual impairment is a problem that affects more than 2.2 billion individuals in the world, with about 43 million people being blind and more than 295 million having moderate to severe vision impairment, which severely restricts free movement and access to environmental information. Conventional devices like white canes and Braille devices give little contextual information and are not real-time interpretative. This paper aims to solve these issues by introducing a real-time wearable assistive system that uses machine learning to integrate computer vision, deep learning, optical character recognition (OCR), and speech processing technologies in a modular client-server system. A wearable camera constantly captures the surrounding and processes it with the help of the YOLOv8 deep learning-based object detection model that allows recognizing surrounding objects with high precision and low latency. Textual data is read in the identified areas by an OCR module, which enables the identification of signs, labels, and printed documents. NLP is used to understand voice queries by the user and contextualize visual responses, but Speech-to-Text (STT) converts voice input into machine-readable text. This processed information is then converted into audio feedback by a Text-to-Speech (TTS) engine to enable hands-free interaction. The system architecture facilitates real time cloud-based inference and modular scalability, which increase environmental awareness, accessibility, and independent navigation of users with visual impairments.