A Gamified Web Platform for the Automated Diagnosis of Childhood Phonological and Phonetic Disorders through Deep Learning

This paper presents a gamified web platform for the automated diagnosis of children's phonetic–phonological disorders. The system integrates deep learning models with acoustic representations extracted using Wav2Vec2 and structured linguistic coding. It was evaluated on a clinical corpus of over 700 recordings, using cross-validation and a comparison between seven classification models. The model based on deep dense networks achieved an accuracy of 83.57%, exceeding the commonly accepted clinical threshold. In addition, the system reduced the evaluation time by 49.6% compared to the traditional method. The system was preliminarily evaluated using speech data collected from 10 children, focusing on technical feasibility and performance trends rather than definitive clinical validation. While the obtained results show promising classification accuracy, they should be interpreted as an initial proof of concept. The results support its applicability as an objective, accessible, and scalable tool in clinical and educational contexts.