BrailleMail: Voice-Controlled Email Assistant

Visually impaired individuals face significant challenges in accessing conventional email platforms that rely heavily on graphical interfaces and manual navigation. This study proposes BrailleMail, a robust voice-enabled email system designed to enhance digital communication accessibility for the visually impaired. Leveraging speech-to-text (STT), text-to-speech (TTS), and keyword spotting technologies, BrailleMail facilitates seamless email interactions through intuitive voice commands. A custom lightweight convolutional neural network (CNN) architecture was trained and evaluated using the Google Speech Commands Dataset (Version 2), which contains over 105,000 labeled one-second voice clips across diverse command categories. Experimental evaluation on a subset of 10 core voice commands revealed a high classification accuracy of 96.6%, with an F1-score of 0.967, demonstrating the model's robustness in recognizing user intents even under noisy conditions. A detailed confusion matrix analysis confirmed minimal misclassifications and strong generalization across varied speaker accents. The system supports command execution such as reading, composing, and navigating emails purely via voice, significantly improving accessibility and independence for visually impaired users. Future enhancements include expanding multilingual support and integrating with mainstream email services like Gmail and Outlook..