English abstract
Parkinson’s disease (PD) is a neurodegenerative disorder that affects movement, speech,
and other functions. Dysarthria is a speech disorder often associated with PD. It can reduce
intelligibility, making it difficult for people with PD to communicate. In recent years,
there has been a growing interest in the use of machine learning (ML) and artificial
intelligence (AI) to diagnose this disorder. This study proposes an automated system for
classifying the severity of dysarthria in patients with PD using a combination of speech
signal analysis, ML algorithms, and voting techniques. This study uses a variety of
features extracted from the speech signal, including statistical signal characterization
(SSC), Mel-frequency cepstral coefficients (MFCCs), and discrete packet wavelet
decomposition (DPWD). These features are used to train ML models, such as support
vector machines (SVM), k-nearest neighbors (KNN), neural networks (NN), and
convolutional neural networks (CNN). Two versions of the system were implemented:
one using the National Technical Institute of the Deaf (NTID) scale, which classifies the
severity of dysarthria into five levels, and a simpler version that classifies the severity into
three levels: mild, moderate, and severe. Several approaches were explored to improve
the test accuracy, including segmenting the subjects, and voting on different levels. The
results of this project can assess the treatment of speech disorders in patients with PD and
improve the success of the speech therapies. The results of the evaluation showed that the
system was able to classify the severity of dysarthria using DPWD and hard voting
between KNN, SVM, and NN with an accuracy of 90.5% using Segment-Subject-Model
voting on the data collected by the hospital of the university of Kiel, Germany.
