Apppublic

vpatel2398/parkinsons-voice-analysis

sourceHugging Facemitupdated 4mo agoView on Hugging Face

0likes

App README

Parkinson's Voice Analysis Demo

Live demo of an acoustic-feature-based Parkinson's disease classifier.

Upload or record a sustained /a/ vowel, and the system extracts 16 voice quality features using Praat (jitter, shimmer, HNR, F0 statistics), then runs a Random Forest classifier trained on the UCI Parkinson's voice dataset with subject-grouped 5-fold cross-validation.

Performance

Metric	Random Forest
Accuracy (subject-grouped CV)	75.3% ± 8.0%
ROC-AUC	78.9% ± 11.9%
PR-AUC	90.6% ± 8.9%

Honest disclaimers

—Not a medical device. Cannot diagnose Parkinson's disease.
—Trained on 31 subjects (UCI dataset) — limited demographic generalization.
—Educational portfolio project, not clinically validated.

Source code

Full methodology, training pipeline, and documentation: github.com/vpatel2398/Parkinson-Disease-Detection-with-Audio