Automatic Speech Recognition (ASR) systems are essential across a wide range of applications, from voice assistants to transcription services. Despite significant advancements, challenges persist, particularly in accurately recognizing speech in noisy environments and identifying voice disorders. This paper presents a comprehensive approach that integrates contemporary techniques for error detection and correction in ASR with a deep learning-based model for detecting voice disorders. Our proposed method utilizes both decoder-based and non-decoder-based features, enhanced by linguistic context, to improve accuracy and usability. Additionally, the system incorporates personalized communication support features aimed at assisting individuals with speech difficulties. By leveraging advanced machine learning techniques to analyze vocal characteristics, our model not only reduces Word Error Rate (WER) in ASR outputs but also aids in the accurate identification of various voice disorders. This research highlights the potential for improving the quality of life for individuals with voice disorders through effective diagnostic and therapeutic solutions.