Reconhecimento de padrões em medidas acústicas para identificação de patologias da laringe

Abstract

The diagnosis of laryngeal pathologies has been performed predominantly in a qualitative way, being obtained through an invasive and uncomfortable examination to the patient. Through digital signal processing techniques (applied to the sustained vowel / a /), so-called acoustic measures such as jitter (or disturbance of the fundamental frequency of the voice), shimmer (disturbance of the amplitude of the voice signal throughout vocalization time) and the relation between harmonic and non-harmonic energy can be obtained to quantify the abnormal functioning of the larynx in order to obtain more precise results. From the statistical evaluation of these measures, using the artificial neural networks, can, then, define a model for the identification of the pathologies that affect the structures of the larynx.