Artificial intelligence approach for detecting pathological voice

Chronic diseases can severely impair the quality of life of patients and their families and take an astronomical toll on our health care system. Early detection of the illness may solve these problems as well as lead to more cures. The purpose of this study is to detect pathological voice with artificial intelligence, as voice changes may be a precursor to a disease such as laryngeal cancer. Machine learning and deep learning algorithms used in this study include support vector machine, extreme gradient boosting, logistic regression, lightGBM, one-dimensional, and two-dimensional convolutional neural networks. Data files consist of 4-second /a:/ phonation collected from 180 subjects. Pre-processing and feature engineering, such as the PRAAT method is executed to compensate for the limited number of samples by extracting well-established features from voice signals. In a binary classification task, all algorithms showed a meaningful degree of accuracy with the highest of over 90% from a one-dimensional convolutional neural network. Our study proves that artificial intelligence may serve as an early and non-invasive diagnostic tool. Further studies are needed to establish a multi-class classification model, for example, differentiating a certain type from benign ones.