Classificador neural de vozes rugosas

Abstract

The human voice is one of the main means of communication, it is also the main tool of work of several professionals. Considering the difficulties that exist in voice assessment diagnoses and the need for voice care, the present study proposes an alternative and non-invasive method to assist professionals in the speech therapy area using artificial neural networks to classify human voices with roughness characteristics. The extraction of the characteristics of the signal in time-frequency analysis will be performed through the wavelet packet transform, and the neural network used for the disturbance classification will be multiple layers Perceptron. The developed classifier achieved a high percentage of correct answers, proving to be a promising alternative for future applications as a non-invasive method that assists speech therapy professionals.