Speech Signal Driven Diagnostic Framework for Parkinson ’s Disease using Cepstral Feature Analysis and DWT Transformation

Parkinson’s Disease (PD) is a kind of neurodegenerative disease that would affect motor control and speech. Only when diagnosed early the progression of PD could be delayed. However, more traditional diagnostic methods lack sensitivity at various stages. This proposed research provides a unique, non-invasive and computationally efficient framework which can be used for predicting PD very early. The proposed framework uses speech signal and its features like cepstral feature extraction and with Discrete Wavelet Transform (DWT). The framework also integrates Ant Colony Optimization (ACO) for feature selection and Adaptive Support Vector Machine for classification. Initially, the method starts with transforming the raw speech signal using DWT. This is done to isolate relevant time-frequency components and to reduce the noise. The key features like MFCC (Mel Frequency Cepstral Coefficients), LPC (Linear Predictive Coding). LPCC, PLP and RASTA-CEPS are extracted from the speech signal to track the speech dynamics. ACO a well-known optimizing algorithm is used to select optimal feature subsets. The optimized features are classified using Adaptive Support Vector Machine (ASVM) which tunes the kernel parameters to evolve feature space. The UCI Parkinson’s Dataset was used and validated using real time recorded speech signals. The proposed method and model achieved accuracy of 94.6%, precision 93.2%, recall of 95.1% and F1 score of 94.1%, outperforming conventional SVM and other classifiers in early PD prediction. The proposed framework proves as a highly effective approach for non-invasive diagnosis of PD as it combine naturally inspired optimization with speech signal processing. A GUI based framework enhances its usability for telemedicine and point-of-care diagnostics.