CPR-SAT: A context-aware pictogram recommendation system – a speech assistive technology

CPR-SAT is an open-source assistive communication software developed for individuals with speech impairments using Augmentative and Alternative Communication (AAC). The system converts spoken and typed input including colloquial, shortened, and phonetically distorted words into structured pictogram sequences with speech synthesis. While recent AAC research has increasingly explored deep learning–based pictogram prediction models that rely on large-scale pretrained representations, CPR-SAT adopts a modular and interpretable architecture that explicitly separates linguistic normalization, semantic retrieval, and contextual re-ranking. The normalization pipeline integrates rule-based correction, dataset-driven similarity matching, and lightweight AI fallback under noisy input. Context adaptation is formulated as a retrieval re-ranking mechanism using time and location metadata. Contextual re-ranking improved Precision from 0.81 to 0.98 with a mean absolute error reduction of 0.09, and achieving 90.76% Top-1 and 96.37% Top-9 pictogram prediction accuracy. A four-week study showed 50% faster communication and 57% fewer interaction taps, demonstrating practical usability gains.