Chatphasia: a personalized end-to-end system for aphasia therapy

This paper introduces an AI-powered system designed to revolutionize therapy for aphasia patients by enhancing accessibility and personalization. The solution integrates a mobile application for patients to perform word-retrieval tasks and a web dashboard for practitioners to monitor progress and tailor therapy. Key components include a fine-tuned Automatic Speech Recognition (ASR) model optimized for Singaporean aphasic speech and a Large Language Model (LLM)-driven cue generation framework based on hierarchical techniques used in speech therapy. The combination of these components forges a comprehensive tool for speech rehabilitation, demonstrating significant improvements in usability, transcription accuracy, and therapy outcomes.