Deep Learning Powered Multimodal SMS Processing and Speech Conversion

This project focuses on addressing the challenge of multilingual SMS classification and speech synthesis by incorporating advanced deep learning models and secure delivery mechanisms. Traditional systems often struggle with real-time processing, language identification, and accurate message classification, particularly in applications that support diverse language inputs and require accessibility for users with reading difficulties. Drawing from previous work that employed deep learning for text classification, this system improves upon existing limitations by combining clustering algorithms such as K-means, DBSCAN, and Agglomerative Clustering with classification models like MAML, ULMFit, Linear SVC, Gaussian Naïve Bayes, and Bagging with Decision Tree. The system also integrates real-time speech synthesis for classified messages while also ensuring secure audio delivery in private spaces. Designed for scalability and crosslingual adaptability, the system achieves classification accuracy of up to 94%, with F1-scores exceeding 0.92 across various categories. This demonstrates its capability to enhance text message accessibility while maintaining data confidentiality and processing efficiency, especially in multilingual communication scenarios.