Speech disorders, particularly dysarthria, are speech impairments resulting from neurological conditions that affect muscle control needed for clear speech. Detecting these disorders at an early stage is crucial for proper intervention, yet manual diagnosis can be time-consuming and challenging. This project aims to automate the classification of speech disorders using deep learning models, specifically convolutional neural networks (CNN) combined with recurrent neural networks (RNNs) like LSTM (Long Short-Term Memory) and GRU (Gated Recurrent Units), as well as the Wave2vec model. The dataset utilized for training the models is the "noise-reduced-uaspeech-dysarthria-dataset," which contains speech samples with varying levels of dysarthria severity. The deep learning models are trained to extract temporal and spectral features from these speech samples, enabling accurate classification. This project involves developing a web application that allows users to upload speech recordings for automated classification of dysarthria, providing quick and reliable results. The primary goal of this project is to build a robust model that can identify different speech disorders based on speech features, assisting healthcare professionals in early detection and diagnosis.

Keywords: Speech disorder, dysarthria, deep learning, convolutional neural network, recurrent neural network, LSTM, GRU, Wave2vec, classification, healthcare.