Lopes Leonardo Wanderley, de Abreu Samuel Ribeiro
Communication Science Disorders Department, Graduate Program in Decision Models and Health, Universidade Federal da Paraíba - UFPB, João Pessoa, PB, Brazil.
Graduate Program in Decision Models and Health, Universidade Federal da Paraíba - UFPB, João Pessoa, PB, Brazil.
J Voice. 2024 May 8. doi: 10.1016/j.jvoice.2024.04.021.
To identify the accuracy and cut-off values of the cepstral peak prominence (CPP) and cepstral peak prominence-smoothed (CPPS) obtained from different speech tasks, to identify dysphonic voices in Brazilian Portuguese speakers, and to verify the correlation between these measures and the overall severity of dysphonia (OS).
In a study with 376 subjects-277 with dysphonia and 99 controls-we recorded four speech tasks and assessed OS with a visual analog scale. We extracted CPP and CPPS from these recordings and analyzed them using receiver operating characteristic curves to determine cut-off values and other performance metrics (area under curve, accuracy, sensitivity, specificity, predictive values, and likelihood ratios).
CPP values below 28.15 dB (sustained vowel [Ɛ]), 28.77 dB (sustained vowel [a]), 28.58 dB (all Consensus Auditory-Perceptual Evaluation of Voice (CAPE-V) sentences), and CPPS values below 16.42 dB (sustained vowel [Ɛ]), 17.02 dB (sustained vowel [a]), and 11.30 dB (all CAPE-V sentences) are indicative of the presence of dysphonia. CPP, CPP, CPP, CPP, CPPS, CPPS, CPPS, CPPS e CPPS, and CPPS can explain the variability of OS from 19% to 57.9%.
The cut-off values for CPP, CPP CPP, CPPS CPPS and CPPS obtained from Praat software can be used as a reference for the clinical voice assessment of Brazilian Portuguese speakers in the respective speech tasks. Analysis of the correlation coefficients of the cepstral measures investigated provided us with a predictive model of OS perception based on each measure/speech task.
