AI Based Automatic Speech Assessment and Personalized Therapy for Aphasia Patients

Abstract

Aphasia is a neurological condition that affects an individual's ability to speak and understand language. Accurate analysis and evaluation of patient speech are essential for effective rehabilitation. In this study, an artificial intelligencebased speech therapy system is proposed to support aphasia rehabilitation. Speech samples are collected through structured tasks and processed using preprocessing techniques such as noise removal, silence removal, and normalization to enhance audio quality. Feature extraction is performed using Mel Frequency Cepstral Coefficients (MFCC), pitch, and signal energy to capture speech characteristics. Automatic Speech Recognition (ASR) is employed to convert speech into text, while acoustic and language models are used to manage variations in pronunciation and word sequences. Furthermore, Natural Language Processing (NLP) techniques are applied to identify missing words, grammatical errors, and pronunciation mistakes. The system provides rulebased feedback by mapping detected errors to appropriate speech therapy exercises. In addition, longitudinal analysis is conducted to evaluate patient performance across multiple sessions and track progress over time. The proposed system improves the efficiency and accuracy of speech evaluation in aphasia rehabilitation, reduces manual effort, and provides structured feedback for continuous patient improvement.