Speech organs are very susceptible to several types of pathologies, which may harm voice production. Several techniques have been traditionally used to detect these pathologies. However, they present drawbacks concerning the accuracy and the comfort of patients during application. Moreover, results obtained by computing techniques have not yet matured to a reliable tool for application in clinics. In this research, a classification approach based on a method not previously employed in classification of vocal tract diseases is proposed. It is based on Prediction by Partial Matching (PPM), which uses acoustical and temporal features to feed models. It were obtained very promising results in the presence or absence of pathologies (at least 92%). With regard to pathology discrimination, preliminary results confirmed that PPM is a high potential technique for voice pathology classification, although its clinical application for the diagnosis of voice pathologies still needs deeper investigation.
Keywords-speech pathologies; prediction by partial matching; acoustical and temporal features;Um dos métodos mais eficazes de compressão de dados é a Predição por Casamento Parcial (Prediction by Partial Matching -PPM) [18]. Seu princípio de funcionamento será descrito na Seção II.B. Contudo, deve-se considerar que bons a Diferentes diagnósticos podem ser dados por diferentes profissionais ou, até mesmo, pelo mesmo profissional, em ocasiões diferentes.