The Relationship Between Cepstral Peak Prominence and Pitch Strength in Complex Dysphonia

Objective: Objective estimates of voice quality have included acoustic measures, such as the cepstral peak prominence (CPP), and psychoacoustic measures, such as pitch strength (PS). Both measures are sensitive to periodicity, both are correlated with perceived dysphonia severity, and PS measures correlate strongly with perceived breathiness and less well with perceived roughness. Because these measures do not depend upon the identification of cycles of vibration in the acoustic waveform, they can be used as indices of voice quality in highly aperiodic voice samples. This presentation explores the relationship between these measurements in a set of complex voice samples that covaried in breathiness, roughness, and strain.

Methodology

Four voice experts selected from various voice databases sustained vowel samples that had covarying levels of breathiness, roughness, and strain. Samples represented 3 levels of severity (mild, moderate, severe) of breathiness x 3 severity levels of roughness x 3 severity levels of strain to result in 27 possible levels of covarying voice quality. Vowel segments were analyzed in terms of CPP (in Praat and ADSV) and PS (using Aud-SWIPE, computed in MATLAB). Measures of CPP F0 and pitch height (PH) also were obtained to investigate frequencies (in Hz) corresponding to CPP and PS.

Results

1. CPP and PS computed from the covarying voice samples were strongly correlated (r values > 0.90) and with vocal severity (r values > 0.75).
2. F0 values corresponding to the indices of CPP and PS were poorly correlated.

Conclusions
These results demonstrate that measures of CPP and PS are highly correlated, likely reflecting their common dependence on strength of periodicity (e.g., both CPP and PS are highly correlated with autocorrelation). CPP and PS both correlate well with ranked severity, even though the frequency locations at which these magnitudes are measured differ considerably. This result supports the premise that, unlike time-based perturbation measures, measures such as CPP and PS are strong predictors of dysphonia severity without depending upon the identification of cycle boundaries and period or F0 estimates. Though CPP and PS are highly correlated, differences in these approaches may make certain methods more beneficial for certain types of dysphonia.