TOWARDS THE EARLY DETECTION OF SPEECH DISORDERS IN CHILDREN APPLYING MANIFOLD LEARNING TECHNIQUES: PRELIMINARY RESULTS

Speech disorders in children can significantly hinder their social interaction, academic performance, and overall development. Early detection is crucial for effective intervention, yet current diagnostic methods often overlook the potential of video-based analysis. This study explores the application of Principal Component Analysis (PCA) and Autoencoder (AE) techniques to analyse video data, focussing on facial movements during speech to facilitate the clustering of children with distinctive speech features. Using a database of 60 video recordings, embeddings were generated and evaluated on the basis of their ability to reconstruct the original data, the results visualised through t-SNE. PCA demonstrated superior performance with a Mean Squared Error (MSE) of 0.0023 for 78 dimensions, while AE achieved its lowest MSE of 0.0093 with 15 dimensions. In particular, embeddings with lower MSE showed better clustering tendencies. This study highlights the potential of integrating video-based analysis into machine learning frameworks to improve the accuracy and depth of speech disorder diagnostics.