ABSTRACT
Dysarthria is a voice disorder in which the muscles responsible for speech production and articulation are weak or the person has difficulties in controlling them. Dysarthria is identified by slurred or collapsed speech that can be unintelligible. The objective of this work is to develop a Computer-Aided Diagnosis (CAD) system to classify the severity of dysarthria. The proposed work helps speech-language pathologists/clinicians to assess the severity of the disorder thereby providing appropriate treatment and therapy. Temporal and spectral features such as short-time energy, energy entropy, zero crossing rate, spectral centroid, spectral roll-off, and spectral flux are extracted from speech utterances in the Nemours database, and the severity is classified using a backend Machine Learning (ML) classifier. Using multi-layer perceptron as a backend classifier, a significant improvement in the performance metrics of multiclass dysarthric severity assessment is observed. Henceforth, the proposed CAD tool helps toward accurate severity assessment of the type of dysarthria for improving the rehabilitation of dysarthric patients.
