Vocal Biomarkers and Deep Learning Approaches for Early Parkinson’s Disease Detection

Early discovery of Parkinson's disease (PD) remains vital because it enables effective management alongside treatment opportunities. Vocal biomarkers, derived from speech recordings, offer a noninvasive method for early PD diagnosis. In this research the use of various machine learning and deep learning techniques are used to analyze vocal biomarkers for early disease detection. Two datasets are utilized: one consisting of CSV files with extracted features, and the other comprising raw audio files. For the CSV dataset, three detection algorithms known as XGBoost Classifier and Support Vector Machine (SVM) were used and Random Forest—are employed. The raw audio files are processed using Convolutional Neural Networks (CNNs), and VGG 19, to automatically extract and learn relevant features for PD detection. These algorithms are trained and evaluated on features extracted using Mel Frequency Cepstral Coefficients (MFCC), Chroma, and Mel-spectrograms. The result output shows how CNN-MFCC with an accuracy of 90% CNN-Mel with an accuracy of 82.6%, CNNChroma with an accuracy of 56.3% and a pretrained algorithm VGG-19 gave an accuracy of 61%. The performance of each classifier is measured to determine their effectiveness in detecting early-stage PD from the vocal features. The CNN-based approach leverages the powerful feature extraction capabilities of deep learning to analyze the complex patterns within the speech recordings.