Estimação do grau de parâmetros subjetivos vocais aplicando redes neurais artificiais

Abstract

Under the psychosocial and cultural view, the voice brings with it the need for human communication, carrying the identity of each individual, being as personal as fingerprints, being one of the strongest expressions of the human personality. Therefore, the eminence of any vocal disorder can cause several aggravating for the person in question. This work encompasses both the field of engineering and speech therapy, bringing a computational tool, developed through artificial neural networks and wavelet packet transform that is capable of assisting professionals who study vocal disorders through a non-invasive method, such as perceptual-auditory, which uses the perception of the evaluators to classify subjective parameters of the voice, such as roughness, breathiness and tension. The computational tool developed will help both the evaluators to analyze themselves in their audible perceptions, as well as to shorten the evaluation time, bringing the classification of the predominance of the subjective parameters of the analogue visual scale, as well as the estimation of these parameters in percentages. Such elaborated computational model proved to be efficient in the analysis of the results, presenting precision rates above 95% in the classification and estimation of the subjective parameters addressed in the study.