Automated Detection of Parkinson's Disease Using Convolutional Neural Networks and Synthetic Spectral Image Features

This release contains the complete, reproducible implementation of the real-data experiment for the paper: Domínguez-Monterroza A., Mateos-Caballero A., Jiménez-Martín A. Automated Detection of Parkinson's Disease from Voice Recordings Using Convolutional Neural Networks and Synthetic Spectral Image Features, 2026 (In Review) Acknowledgements: This work was funded by grants PID2021-122209OB-C31 and RED2022-134540-T from MICIU/AEI/10.13039/501100011033 Included Components 1. Full Source Code This release provides all scripts required to: Load raw audio signals (CSV format) Extract MFCC spectral images using librosa Build a Convolutional Neural Network (CNN) classifier in TensorFlow/Keras Perform 10-fold stratified cross-validation Compute metrics: Accuracy, F1, Precision, Recall, ROC AUC Generate visualization of spectral differences between control and PD subjects Scripts included: 01_extract_mfcc.py 02_build_cnn.py 03_cross_validation.py 04_visualization.py full_pipeline.py Dataset source (PC-GITA) The real voice dataset comes from: J. R. Orozco-Arroyave, J. D. Arias-Londoño, J. F. Vargas-Bonilla,M. C. Gonzalez-Rátiva, and E. Nöth, New Spanish speech corpus database for the analysis of people suffering from Parkinson's disease, Proc. 9th Int. Conf. Language Resources and Evaluation, 2014. The dataset includes: 100 Colombian speakers 50 PD 50 healthy controls Each participant phonated /a/ three times → 300 total recordings (150 PD, 150 HC). Synthetic data generation Dataset source: M. Rey-Paredes, C. J. Pérez, A. Mateos-Caballero,Time Series Classification of Raw Voice Waveforms for Parkinson's Disease Detection Using Generative Adversarial Network-Driven Data Augmentation, IEEE Open Journal of the Computer Society, 2025.