Acoustic and Perceptual Classification of Within-sample Normal, Intermittently Dysphonic, and Consistently Dysphonic Voice Types

Abstract

Intermittent dysphonia within an utterance is common, but presents difficulty for both perceptual and objective voice evaluation. This study examined the ability of measures from the within-sample cepstral peak prominence (CPP) distribution to differentiate normal voices from intermittently and consistently dysphonic voices. Exploratory design. Sixty samples of the sentence "We were away a year ago" were classified as normal, intermittently dysphonic, or consistently dysphonic by four judges. Measures of CPP within each sample were obtained, and further analysis with examined CPP distribution variability and patterns of CPP outliers. Whereas the mean CPP was the strongest single discriminator among the three voice types, the normal and intermittent dysphonia groups were not significantly different on CPP distribution skewness and measures of CPP distribution outliers. Both the normal and intermittently dysphonic voices differed significantly from the consistently dysphonic samples on these variables. A combination of measures of the CPP distribution was effective for a linear prediction of percent dysphonia duration for the speech samples (r = 0.825; rho = 0.81). The CPP standard deviation significantly improved the use of the mean CPP in discriminant function analyses and also the classification of the intermittently dysphonic voices. Auditory-perceptual judgment of dysphonic segments and the typically robust acoustic measurement of mean CPP are both ineffective for classifying intermittently dysphonic voices. However, dysphonia duration may be effectively predicted via measures of the CPP distribution, and acoustic classification of dysphonic voice types via cepstral methods may be improved with an analysis of the CPP distribution across an utterance.