Prediction of vocal severity within and across voice types

Fifty-one subjects representing diverse laryngeal etiologies recorded /a/ and /i/ to provide a study sample of 102 vowel sounds. Listeners categorized each vowel on the basis of four voice types (normal, breathy, hoarse, unclassified) and evaluated the degree of vocal abnormality on a 7-point scale. In addition to spectrographic noise (SN) classification, several acoustic measures based on period variability were entered into a multiple regression analysis for the prediction of vocal severity across and within voice types. In general, spectrographic noise and curvilinear derivatives of the period standard deviation (PSD) provided the best predictions of disorder severity. Different variables were the major predictors for different voice types. Several variables used in previous studies were inefficient as predictors of severity.