Comparative Analysis of Pitch Detection Algorithms for Machine Learning Supported Parkinson’s Disease Diagnosis

Abstract

Parkinson’s disease (PD) is a progressive neurodegenerative disorder characterized by motor and non-motor symptoms, including hypokinetic dysarthria (HD), a speech disorder affecting prosody. Early detection of PD through speech analysis offers a promising, non-invasive diagnostic approach. This study evaluates five pitch detection algorithms—PRAAT, YIN, PYIN, RAPT, and SWIPE’—to extract fundamental frequency-based features from the PARCZ speech database. The extracted features, including relative F 0, standard deviation and various jitter measures, are used to train and evaluate three binary classifiers: Logistic Regression (LR), Support Vector Machine (SVM), and Random Forest (RF). The classifiers are optimized using a stratified crossvalidation approach, with balanced accuracy as the primary metric. Results indicate that while pitch-based features alone are insufficient for clinically accurate PD diagnosis, certain classifiers and pitch detection methods show potential in aiding early detection. Future work should incorporate a broader set of speech parameters to enhance diagnostic precision.