Reproducible Machine Learning-based Voice Pathology Detection: Introducing the Pitch Difference Feature

TL;DR

This work presents a reproducible ML framework for detecting voice pathology using the Saarbrücken Voice Database, introducing two novel features—Pitch Difference and NaN—to improve discrimination. The authors perform extensive feature subset exploration across six traditional ML algorithms, address class imbalance with k-means SMOTE, and enforce strict data handling to prevent leakage, including sex-aware modeling. They evaluate using low-bias metrics (MCC, UAR) and provide detailed reproducibility artifacts (code, data metadata, and checksums) following REFORMS guidelines. The study demonstrates robust detection performance on sustained /a:/ vowels and underscores the need for reproducibility and careful data curation in clinical voice pathology research.

Abstract

Purpose: We introduce a novel methodology for voice pathology detection using the publicly available Saarbrücken Voice Database (SVD) and a robust feature set combining commonly used acoustic handcrafted features with two novel ones: pitch difference (relative variation in fundamental frequency) and NaN feature (failed fundamental frequency estimation). Methods: We evaluate six machine learning (ML) algorithms -- support vector machine, k-nearest neighbors, naive Bayes, decision tree, random forest, and AdaBoost -- using grid search for feasible hyperparameters and 20480 different feature subsets. Top 1000 classification models -- feature subset combinations for each ML algorithm are validated with repeated stratified cross-validation. To address class imbalance, we apply K-Means SMOTE to augment the training data. Results: Our approach achieves 85.61%, 84.69% and 85.22% unweighted average recall (UAR) for females, males and combined results respectively. We intentionally omit accuracy as it is a highly biased metric for imbalanced data. Conclusion: Our study demonstrates that by following the proposed methodology and feature engineering, there is a potential in detection of various voice pathologies using ML models applied to the simplest vocal task, a sustained utterance of the vowel /a:/. To enable easier use of our methodology and to support our claims, we provide a publicly available GitHub repository with DOI 10.5281/zenodo.13771573. Finally, we provide a REFORMS checklist to enhance readability, reproducibility and justification of our approach

Figures (7)

• Figure 1: Flow diagram of the proposed methodology
• Figure 2: Healthy female subject ID 1
• Figure 3: Female subject ID 1, suffering with functional dysphonia
• Figure 4: Age distribution of healthy and pathological male and female subjects
• Figure 5: Preprocessing before feature extraction