Fourier Model-Based Analysis of LP Residual for Diagnosing Parkinson’s Disease Using Speech

In this investigation, two new features, residual harmonic amplitude (RHA) and residual harmonic frequency (RHF) are proposed using Fourier model-based analysis of the linear prediction (LP) residual signal for diagnosing Parkinson’s disease (PD) from the speech signal. The excitation source information is represented by the LP residual of the speech signal. The LP analysis determines the LP residual of speech by performing inverse filtering operations on the speech signal. The proposed RHA and RHF features are extracted from the LP residual using Fourier model analysis. The probability density function (PDF) plots and box plots are used to examine the importance of the proposed features in classifying PD patients and healthy people. The effectiveness of the proposed features for detecting PD is evaluated using the support vector machine (SVM) classifier. The results show that the proposed RHA and RHF features are effective for detecting PD. The proposed features give comparable performance with linear predictive coefficient (LPC) features for discriminating between PD-affected and healthy speech.