Related Experiment Video
Updated: Oct 9, 2025
04:04
Asthma Detection Research Based on Voice Signal Processing and Machine Learning
Published on: July 22, 2025
515
Development of a machine-learning based voice disorder screening tool
Jonathan Reid1, Preet Parmar2, Tyler Lund3
1Division of Otolaryngology-Head and Neck Surgery, Department of Surgery, Faculty of Medicine and Dentistry, University of Alberta, Edmonton, AB, Canada.
American Journal of Otolaryngology
|December 19, 2021
Summary
A new machine learning algorithm (MLA) effectively screens for voice disorders using audio samples. This AI tool shows high accuracy in detecting pathological voices, offering a promising solution for early diagnosis and management.
Area of Science:
- Otolaryngology
- Artificial Intelligence
- Biomedical Engineering
Background:
- Effective voice disorder management relies on early recognition and referral.
- Limited access to specialists and primary care awareness hinder timely diagnosis.
- A need exists for advanced tools to improve voice pathology screening.
Purpose of the Study:
- To design and validate a machine learning algorithm (MLA) for detecting pathological voices.
- To assess the MLA's performance using a combination of convolutional neural networks (CNN) and Support Vector Machines (SVM).
Main Methods:
- Developed a MLA by converting audio samples into spectrograms for input into a VGG19 CNN.
- Classified features using a Support Vector Machine (SVM) binary linear classifier.
- Trained and tested the MLA on the Saarbrucken Voice Database (SVD) and externally validated with clinical samples.
Main Results:
- The MLA achieved 98.5% sensitivity, 97.1% specificity, and 97.8% accuracy on SVD samples.
- External validation demonstrated 100% sensitivity, 96.3% specificity, and 96.7% accuracy on clinical samples.
Conclusions:
- A MLA utilizing simple audio input can identify diverse vocal pathologies with high accuracy.
- This algorithm presents a promising potential as a voice pathology screening tool.
