NeuroVox uses vocal biomarkers and machine learning to detect Parkinson's disease in its earliest stages. By training models like Random Forest and KNN on speech data, the platform achieves over 94% test accuracy and 100% recall—supporting physicians with data-driven diagnostics through a user-friendly interface.
The Diagnostic Challenge
NeuroVox is a diagnostic tool built to revolutionize how Parkinson's disease is detected—by using something as simple and accessible as a person's voice. Parkinson's is a progressive neurodegenerative disorder that impacts motor function, speech, and quality of life. Yet in its early stages, it's notoriously hard to detect. Minor vocal changes—such as reduced pitch variation or vocal tremors—often go unnoticed by the human ear and standard clinical assessments.
Our Machine Learning Approach
NeuroVox bridges this diagnostic gap using machine learning models trained on a dataset of voice recordings. By analyzing acoustic features like jitter, shimmer, harmonic-to-noise ratio, and fundamental frequency, our models can detect early-stage Parkinson's with high precision. Our workflow includes full data preprocessing (duplicate removal, outlier detection, MinMax normalization), feature engineering, and model tuning using techniques like cross-validation, SMOTE balancing, and grid search.
Performance Results
Across six classification models—Logistic Regression, SVM, Decision Tree, Random Forest, K-Nearest Neighbors (KNN), and Neural Networks—we found that Random Forest and KNN performed the best, each achieving 94.87% test accuracy with 100% recall and F1-scores of 96.97. This means all actual Parkinson's cases were detected, which is critical in any medical application to avoid false negatives. We evaluated all models using standard metrics: accuracy, precision, recall, F1-score, and cross-validation performance.
Clinical Integration
User-Friendly Interface: The entire system is deployed as a user-friendly interface where patients or clinicians can upload voice samples and receive real-time feedback. Our design prioritizes accessibility for integration into telemedicine platforms, making early-stage diagnostics more available to underserved regions without neurologists.
Supporting Clinical Judgment: NeuroVox does not aim to replace physicians—it enhances clinical judgment by offering a powerful second opinion grounded in data. With future iterations, we aim to expand the platform to detect other neurodegenerative diseases using vocal and motion biomarkers.
Research & Development
To explore our full technical report and research paper: Technical Documentation
Impact Vision
NeuroVox represents the democratization of neurological diagnostics, making early detection accessible worldwide through the simple act of speaking. By leveraging the power of machine learning and vocal biomarkers, we're creating pathways for earlier intervention and better patient outcomes in neurodegenerative diseases.