Hybrid voice spectrogram-chromogram based deep learning (HVSC-DL) model for the detection of Parkinson's disease

Abstract

Parkinson's disease is one of the serious neurological disorders that restricts the life quality of individuals significantly. The changes in sound signals contain important clues for detecting the disease at an early stage. In this study, a newly collected Parkinson's voice dataset is introduced, and preliminary results with classical machine learning and the proposed three alternative deep learning models are presented comparatively. We observed that our proposed two-channel Hybrid Voice Spectrogram-Chromogram based Deep Learning Model (HVSC-DL) with the patient-healthy classification accuracy rates of 96.6%, 92.9% and 94.5% on /a/, /o/ and /i/ sounds respectively, showed superior performance compared to pure tone chromogram and spectrogram based models.