The usefulness of multi voice evaluation: Development of a model for predicting a degree of dysphonia

Abstract

The purposes of this study were (1) to analyze the usefulness of self-report questionnaires, acoustic analysis, and auditory perceptual assessment for screening voice problems; and (2) to develop a new model for predicting a comprehensive voice severity using multi-assessment. A total of 306 voice samples were analyzed in this study (typical group, n=72; dysphonia group, n=234). We performed a receiver operating characteristic analysis to determine the cutoff values of auditory perceptual assessments (visual analog scale), acoustic parameters (spectral- and cepstral-based analyses), and self-report questionnaires for screening voice disorders. We also performed a stepwise multiple regression analysis to verify which combination of parameters (acoustic parameters, and self-report questionnaires) could best predict perceived voice severity. We verified that most of the variables analyzed were useful for voice evaluation, and found to be useful for screening voice problems. Of these, a five-variable model was a useful to predict perceived voice severity (mean R2=.807). The five-variable model consisted of acoustic parameters based on cepstral analysis (cepstral peak prominences in connected speech and sustained vowel task, and low versus high-frequency spectral energy ratio in connected speech task) and self-report questionnaires (total score of the Voice Handicap Index, and rumination score of the Voice Catastrophization Index). We verified that most of the variables were useful for screening dysphonia and five-variable model was a useful to predict perceived voice severity. The five-variable model could be used as an objective criterion for predicting voice severity.