Deep Learning Based Web System for the Automated Diagnosis of Phonological-Phonemic Disorders in Infants

Early diagnosis of phonetic and phonological disorders in infants is crucial for their proper linguistic development. This paper presents a web-based system that automates the evaluation process by using deep learning models to analyze infant pronunciation. The system integrates an interactive video game that facilitates the collection of audios in a playful way. The goal is to improve diagnostic accuracy and efficiency, reducing subjectivity and evaluation time compared to traditional methods. The methodology includes the collection of audios from 3–5-year-old children, manual labeling of the data, and the use of deep neural networks to classify speech disorders into omission, distortion, substitution, and correct pronunciation. The results show that the Deep Neural Networks (DNN) model achieved an accuracy of approximately 95%, outperforming other algorithms evaluated. This system promises to be an effective tool to assist therapists in the early detection of phonetic problems, providing a faster and more accurate diagnosis.