Automated Detection and Severity Assessment of Dysarthria using Raw Speech
Kodali Radha, Mohan Bansal
Abstract
Dysarthria is a medical condition that impairs an individual’s ability to speak clearly due to muscle weakness or paralysis. To diagnose and monitor dysarthria severity, this article proposes the use of a deep learning model that utilizes raw speech waveforms. This approach eliminates the need for feature engineering and enhances the model’s ability to handle noise and speech variability. The proposed system was compared to a standard convolutional neural network (CNN) model and was found to perform better in both dysarthria severity classification and dysarthria/healthy control classification tasks. The results indicate that the SincNet model achieved an accuracy of 95.7% and 99.6% in these tasks, respectively. The proposed system can aid clinicians in diagnosing and monitoring dysarthria severity and could have broader applications in speech-related disorders. The study emphasizes the importance of using raw waveform-based models for speech analysis and demonstrates the effectiveness of SincNet in automatic dysarthria/healthy control classification and dysarthria severity assessment.