F. Amato, M. Fasani, G. Raffaelli, V. Cesarini, G. Olmo, N. Di Lorenzo, G. Costantini, G. Saggio

Obesity and Gastro-Esophageal Reflux voice disorders: a Machine Learning approach

Medical Imaging Analysis and Diagnostics Machine Learning

Automatic assessment of speech disorders is a cutting-edge topic in vocal analysis. Recent studies indicated possible connections between eating disorders and voice alterations. In this work, we assessed the influence of obesity and Gastro- Esophageal Reflux Disease (GERD) on voice, being the former a risk factor for the latter. Moreover, we investigated the mutual influence of the diseases working with a consistent set of features. To these aims, we used vocal tests from 92 subjects, with vocal tests consisting of vowel phonation and sentence repetition, and subjects including healthy controls, obese patients, patients with GERD, and obese patients with GERD. Machine Learning models, consisting of Naive Bayes and Support Vector Machine, were successfully employed on extracted features in binary classifications, resulting in 0.86 and 0.82 of accuracies on validation set in scoring the presence of GERD and obesity, respectively. The absence of performance deterioration when moving to the test set denoted a lack of overfitting. As for the tasks and the features employed, the sentence repetition proved to be more effective than the vowel phonation, while Mel Frequency Cepstral Coefficients, Perceptual Linear Prediction Coefficients, Bark Band Energy Coefficients, and noise measures appear to be among the most significant features for the application at hand.

This article is authored also by Synbrain data scientists and collaborators. READ THE FULL ARTICLE