Dysarthric Speech Detection Using Hybrid Models

Dysarthria is a speech disorder caused by weak or poorly coordinated speech-related muscles. Dysarthria can be caused by various factors such as stroke, multiple sclerosis, or cerebral palsy, as well as brain injury or certain medications. The ultimate goal of our paper is to make a suitable and accurate tool for detecting dysarthric speech that can be used in clinical settings for early diagnosis, treatment planning for dysarthric individuals. In this work, the machine learning models namely CNN (Convolutional Neural Network), and hybrid models, such as CNN, combined with LSTM (Long Short-Term Memory), and CNN combined with GRU (Gated Recurrent Unit) are trained on TORGO dataset, to classify the dysarthric speech from non-dysarthric speech. The model's performance is evaluated on the testing data and all of these models produce more than 95% accuracy rate. From this study, we understood that this methodology can be used in developing Automatic Speech Recognition(ASR) for people with dysarthria, which would become an essential technology for many applications.