Validação das medidas cepstrais para a avaliação de distúrbio de voz em falantes do português brasileiro

• Ano de defesa: 2021
• Autor(a) principal: Sousa, Estevão Silvestre da Silva
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal da Paraíba; Brasil; Ciências Exatas e da Saúde; Programa de Pós-Graduação em Modelos de Decisão e Saúde; UFPB
• Palavras-chave em Português: Acústica da fala; Modelos logísticos; Qualidade vocal; Voz; Distúrbios de voz; Fonoaudiologia; Speech acoustics; Logistic models; Vocal quality; Voice; Voice disorders; Speech therapy; CNPQ::CIENCIAS DA SAUDE::SAUDE COLETIVA
• Link de acesso: https://repositorio.ufpb.br/jspui/handle/123456789/21510

The acoustic analysis is a procedure with greater objectivity, non-invasive and that is currently being used commonly in the vocal clinic. The acoustic analysis involves measures capable of quantifying a certain property of the vocal signal. In vocal signals with mild deviation, it is possible to estimate values of the fundamental frequency (f0) by determining the number of sound waves. However, in voice signals with medium or high irregularity, traditional measures, which today are the most used, may have their reliability compromised. In view of this, cepstral measurements have the perspective of determining f0 and producing an estimate of aperiodicity and / or additive noise without the need to identify individual cycle limits. In the acoustic analysis, cepstral measurements are currently considered the most robust measures to quantify voice deviations, however, cut values have not yet been found for Brazilian Portuguese speakers. Thus, the objective of the present study was to build a decision model based on cepstral measures that predict individuals who speak Brazilian Portuguese with and without deviation in vocal quality. This is a descriptive, documentary, cross-sectional and observational study. The sample of this research was constituted by 305 individuals of both sexes, with the report of the otorhinolaryngologist (78 without voice disorder and 227 with vocal complaint and with laryngeal pathology). The measurements were extracted: Cepstral Peak Prominence (CPP), Cepstral Peak Prominence-Smoothed (CPPS) and spectral decline. The discriminative power of cepstral measures was given by means of Logistic Regression models with the variable: voice disorder and the multiple linear regression model was used to analyze the cephalal measures regarding the deviation of vocal quality. In addition, the cutoff point for ceptral measurements for Brazilian Portuguese in different speech tasks was also obtained. It turned out that the reduced logistic regression model for identifying the presence and absence of voice disorder, includes the variables: CPP CAPE-V, CPPS VGE and CAPE-V decline, presenting a performance of 70.7%, with a sensitivity of 85 % and specificity of 53%. In isolation, the variables CPPS VGE and CPPS VGA showed satisfactory performance to discriminate individuals with and without voice disorder, with high specificity values and low sensitivity values. The measures: CPP CAPE-V, CPPS VGE, CPPS VGA and CPPS CAPE-V are able to predict the variability in the intensity of vocal deviation by 64.32%.