Preditores acústicos das atitudes dos ouvintes em relação às vozes disfônicas: um modelo baseado em machine learning

• Ano de defesa: 2024
• Autor(a) principal: Evangelista, Deyverson da Silva
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal da Paraíba; Brasil; Linguística; Programa de Pós-Graduação em Linguística; UFPB
• Palavras-chave em Português: Atitude dos ouvintes; Distúrbios de voz; Disfonia; Voz; Voice; Acoustics; Vocal quality; Attitude of listeners; Voice Disorders; CNPQ::LINGUISTICA, LETRAS E ARTES::LINGUISTICA
• Link de acesso: https://repositorio.ufpb.br/jspui/handle/123456789/33058

OBJECTIVE: To analyze the relationship between acoustic measurements and listeners' attitudes towards dysphonic voices and develop a model based on Machine Learning (ML) to predict the valence of listeners' attitudes towards dysphonic voices. METHODS: The study is retrospective, descriptive, cross-sectional and quantitative. It had 152 Brazilian listeners and 44 vocal samples. Two steps were carried out: selection of vocal samples and removal of acoustic measurements using the Praat software, version 5.3.77h, using the VoxMore script; and listeners' judgment, using the Scale of Judgment of Attitudes Associated with Dysphonic Voices, which encompasses 12 attributes and their transfer was based on three basic dimensions of attitudes: evaluation, potency and activity. The data were analyzed descriptively and inferentially, using association tests, mean comparison, linear regression and ML. RESULTS: We observed the relationship between acoustic measurements, auditory perceptual judgment (JPA), and listeners' attitudes. In JPA, the configurations of general degree, breathiness and roughness stand out in all attitudes, except calm, which behaved differently, in relation only to breathiness. The acoustic measures that were most related to the perception of attitudes were cepstral, frequency and disturbance, showing moderate and strong correlations with the valence in each attitude, as well as differences in the means of the parameters for positive and negative valence. The General Degree, minimum f0 and SNL influence the judgment of listeners' attitudes, according to the regression analysis. Therefore, the greater the vocal deviation and acoustic changes in the vocal signal, the more negative the judgment. From the ML, it was observed that the CCP, Shimmer, maximum f0 and minimum f0 measures are predictive in the listeners' judgment of dysphonic voices. CONCLUSION: Cestral, frequency and disturbance measurements were those that most related to listeners' attitudes towards dysphonic voices, both in relation to valence and attributes. The greater component of noise and vocal irregularity also influences perception. The measures of maximum f0 and minimum f0, CCP and shimmer are predictors in the judgment of listeners' attitudes towards dysphonic voices.