Research · 2023
Parkinson's Detection from Voice
Vocal Tab Transformer that classifies Parkinson's disease from dysphonia features, beating gradient-boosted trees by 1%+ AUC.
≥1%
Higher AUC than gradient-boosted trees
Context
Research collaboration, published in MDPI Biomimetics (2023).
Problem
Parkinson's disease often shows up early in a person's voice. Dysphonia measures extracted from voice recordings form a large, complex tabular feature set, where Gradient-Boosted Decision Trees were the previous state of the art.
Approach
We proposed the Vocal Tab Transformer, a transformer for tabular vocal features, which outperformed Gradient-Boosted Decision Trees by at least 1% AUC with better precision and recall.
Results
- Published in MDPI Biomimetics (2023)
- Outperforms GBDT by ≥1% AUC
Stack
- Transformers
- Tabular deep learning
Have a similar problem?
Tell me what you're working on and I'll tell you honestly whether I can help.