Voice Assistants for Blind Users

Abstract

This research focuses on developing an advanced voice-assisted system to aid visually impaired individuals, with a particular emphasis on Arabic language support. The study addresses the challenges faced by blind users in environmental interaction and mobility by proposing an integrated solution that combines real-time object detection using YOLOv8, intelligent scene description via the multimodal Gemini Pro Vision model, and intuitive Arabic voice interaction. Implemented on a Raspberry Pi platform, the system demonstrates the feasibility of deploying complex AI for real- time multimodal processing on cost-effective embedded hardware. Key contributions include robust Arabic language support, enhanced environmental awareness for users, and the effective fusion of visual perception with voice commands. Preliminary evaluations indicate significant improvements in users' spatial awareness and confidence, highlighting the system's potential to foster greater independence and social inclusion for Arabic-speaking visually impaired individuals. The work also identifies performance bottlenecks and outlines future enhancements, such as NPU integration and advanced dialectal support, to further improve system efficacy.