Signal processing and statistical procedures to identify laryngeal pathologies

Abstract

This work proposes a modular approach, using signal processing techniques and artificial neural networks for diagnosing of glottal conditions related to laryngeal pathologies. While signal processing techniques are used to extract acoustic features from the human voice, artificial neural networks use these features to perform the diagnosis. These features were based on abnormal movement of vocal folds and incomplete closure of glottis. Simple statistical methods, like robust estimators, Mann-Whitney test and principal component analysis were used to improve the percentage of correct classification, allowing up to 82.22% of identification of the glottal conditions using only voice analysis.