Daily Monitoring of Speech Impairment for Early Parkinson's Disease Detection
Al Jizani Mohammed Kadhim Salman, Grigore Stâmâtescu
Abstract
Chronic neurological diseases progress over time, and among these illnesses, Parkinson's Disease (PD) is the most prevalent. PD causes a major issue affecting people's daily activities. It can be diagnosed and treated by clinical care. Early identification and monitoring of people with Parkinson's are necessary to assess the level of health clinic interference and prevent the disease's symptoms from worsening over time. Assessing the changes in the patient's voice, such as progressive stuttering, is essential for a timely PD diagnosis. This paper investigates how the changes in a person's voice are used to detect PD. The work proposes a deep learning-based end-to-end model. The proposed architecture selects 61 features from integrating two different public datasets of PD. It computes measures from a person's voice including 22 measures and simple features of 39 MFCC. The results were achieved from the use of an existing dataset of 195 samples for 31 people after integrating with 39 MFCC from another public dataset with 97% accuracy using the CNN-LSTM architecture.