Hybrid CNN-GRU Model for Predicting Dysarthric Speech Using Deep Learning Approaches

This research study presents the novel method for dysarthria speech detection and classification. In order to create systems that are capable of accurately categorizing speech patterns that are impacted by dysarthria, it is vital to have speech detection and classification capabilities for dysarthria. The condition known as dysarthria, which is characterized by difficulties in articulation, phonation, resonance, and prosody, is a disorder that leads to a reduction in the intelligibility of speech. This method of classification necessitates the development of machine learning or deep learning models, more specifically a combination of Convolutional Neural Networks (CNN) and Gated Recurrent Units (GRU), in order to accomplish the objective of automatically classifying dysarthric speech into categories such as severity or presence/absence of dysarthria. The primary purpose of dysarthria speech categorization is to bring about the development of a reliable instrument that can be used for identifying, categorizing, and managing dysarthric speech. As an added benefit, the classification of dysarthric speech allows for the adaptation of treatment methods and the acquisition of information about the underlying causes of the condition. The individuals who are affected by this speech difficulty are the target audience for this strategy, which ultimately aims to enhance the quality of life of those individual. The study addresses the challenge of classifying dysarthric speech caused by neurological impairments using a Hybrid CNN-GRU model, achieving accuracies of 70.3% (high-severity), 38% (severe-moderate), 34.5% (moderate), and 10.3% (mild dysarthria).