Título: Algoritmo Bio-inspirado em Colônia de Abelhas Aplicado na Seleção de Características para Detecção de Desvios Vocais
Autores: Sousa, Aldeni;Ferreira, Paulo;Vieira, Vinicius;Costa, Silvana;Correia, Suzete
Feature selection is an important step used in several pattern recognition tasks to identify the significant features and discard irrelevant or redundant ones from the original set. In this research, a feature selection algorithm based on binary artificial bee colony with K-NN classifier is used for the discriminative analysis between healthy voices and voices with vocal deviations (breathiness, roughness and strain). The aim is to determine which acoustic measures based on the recurrence quantification analysis are relevant and contribute more in voice signals discrimination. In thirty two tests performed, 88.33% of accuracy was obtained with a reduction of fifteen for seven features in the discrimination between healthy and strained voices, four features and accuracy of 88.33% in the classification between healthy and breathiness and a reduction for five features, with an accuracy of 93.33%, in the discrimination between healthy and rough voices.
Acoustic Analysis;Vocal Deviation;Feature Selection;Binary Artificial Bee Colony.
Código DOI: 10.21528/CBIC2017-65
Artigo em pdf: cbic-paper-65.pdf
Arquivo BibTeX: cbic-paper-65.bib