This study introduces a real-time assistive system aimed at enhancing the mobility and independence of individuals with visual impairments. The proposed solution integrates a convolutional neural network for object detection and a Direct Convolutional Text-to-Speech (DCTTS) model for generating spoken feedback. The system employs a cloud-based server architecture, enabling cross-device compatibility and efficient processing of visual data captured via mobile phone cameras. When placed in a chest pocket with the camera facing outward, the device continuously scans the user's environment, identifying objects such as traffic signals, pedestrians, and obstacles. Based on predefined decision rules, the system determines when guidance is necessary and delivers spoken alerts to the user through synthesized speech. The object detection model achieved a mean average precision of 74.8%, and the overall latency of the end-to-end system remains within acceptable real-time thresholds. This work demonstrates a scalable and practical implementation of AI-driven accessibility technologies and lays the foundation for further development of personalized, voice-enabled assistive solutions.