RNN-GAN Integration for Enhanced Voice-Based Email Accessibility: A Comparative Study for Visually Impaired Users

This paper presents the integration of Recurrent Neural Network-based Generative Adversarial Networks (RNN-GAN) into voice-enabled email access systems for persons with visual impairments. The other neural network architectures havebeen compared including LSTM and GRU. The goal is to enhance spoken command comprehension, information extraction, and contextually relevant response generation. This study uses RNN-LSTM, RNN-GRU and compares with the proposed model (RNN-GAN). RNN-GAN results in higher word accuracy (approximately 70-95%), real-time processing, and robustness to change of speech as compared with the previous models. The challenges involved computational resources needs and the privacy implications. In the era of machine learning, personalization seems to be a significant factor along with enhanced security, multilingual support and finally it has to make sure that model inference is optimised for these low resource environments. This research work helps to make the email more accessible for visually impaired users and thus improve digital inclusion.