Abstract:
This report delineates the conceptualization, design, and implementation of an advanced Speech Defect Detection System, addressing the critical need for effective identification and analysis of speech disorders. Speech defects pose significant challenges in communication, impacting individuals' quality of life and requiring accurate diagnostic methodologies. The prevalence of various speech disorders and the complexity of their identification necessitate innovative solutions to enhance the efficiency and accuracy of diagnostic processes.
To solve this problem, the system harnesses cutting-edge deep learning methodologies, integrating Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks for in-depth analysis of speech patterns. The architectural framework explores the logical, data, and presentation tiers, detailing the deep learning model, user interactions, and database management. User interface design prioritizes clarity and efficiency, optimizing the overall user experience. A discerning approach to technology stack selection ensures alignment with scalability and performance requisites. Additionally, insights from a survey administered to speech pathology professionals provide valuable data on prevalent speech disorders, diagnostic methodologies, and perspectives on technology integration in assessment processes. The report concludes by underscoring the potential benefits of the proposed Speech Defect Detection System, which contributes to the ongoing advancement of assistive technologies within speech pathology.
After conducting an evaluation on SpeechGuard, it was found that the system's machine learning model delivers excellent outcomes. Metrics such as training loss, validation loss, AUC curve, accuracy curve, precision curve, and recall curve observed during the hybrid model training and evaluation indicated excellent results and performance.
