Quantifying the cepstral peak prominence, a measure of dysphonia

Objective The purpose of this study is to establish normative values for the smoothed cepstral peak prominence (CPPS) and its sensitivity and specificity as a measure of dysphonia. Study Design Prospective cohort study. Methods Voice samples of running speech were obtained from 835 patients and 50 volunteers. Eight laryngologists and four speech-language pathologists performed perceptual ratings of the voice samples on the degree of dysphonia/normality using an analog scale. The mean of their perceptual ratings was used as the gold standard for the detection of the presence or absence of dysphonia. CPPS was measured using the CPPS algorithm of Hillenbrand, and the cut-off value for positivity that has the highest sensitivity and specificity for discriminating between normal and severely dysphonia voices was determined based on ROC-curve analysis. Results The cut-off value for normal for CPPS was set at 4.0 or higher, which gave a sensitivity of 92.4%, a specificity of 79%, a positive predictive value of 82.5%, and a negative predictive value of 90.8%. The area under the receiver operating characteristic (ROC) curve was 0.937 (P < 0.05). Conclusions CPPS is a good measure of dysphonia, with the normal value of CPPS (Hillenbrand algorithm) of a running speech sample being defined as a value of 4.0 or higher.