Estimating Pressed and Breathy Phonation From Cepstral and Spectral Measures.

Abstract

Objective: This study aimed to determine (1) how cepstral peak prominence (CPP) and the amplitude difference between the first two spectral harmonics (H₁-H₂) change as a function of using pressed and breathy voice production in vocally healthy individuals and (2) whether CPP and H₁-H₂ can discriminate breathy and pressed phonation from typical phonation.

Method: Acoustic voice recordings were obtained from 50 vocally healthy individuals. Speech tasks included five repetitions of three sustained vowels (/ɑ, i, u/) and one all-voiced sentence ("We were away a year ago") using three voice production conditions (typical, breathy, and pressed). For each production, we calculated average smoothed CPP and H₁-H₂. Mixed ANOVA models and post hoc pairwise comparisons were used to determine how the acoustic measures differed across phonation conditions and whether those differences varied by sex. Multinomial logistic regression models were used to determine whether measures could differentiate the phonation types, controlling for pitch and intensity.

Results: In connected speech, pressed phonation demonstrated higher CPP and lower H₁-H₂ compared with typical phonation, whereas breathy phonation had lower CPP and higher H₁-H₂. Female participants also exhibited higher CPP and lower H₁-H₂ during pressed phonation compared with typical phonation on all sustained vowels. Male participants did not have significant differences in either measure between typical and pressed phonation in any sustained vowel but did have lower CPP/higher H₁-H₂ during breathy phonation. CPP and H₁-H₂ each significantly differentiated pressed from typical phonation in most speech tasks, but neither differentiated pressed from typical phonation in sustained /i/ tasks. CPP differentiated breathy from typical phonation when controlling for intensity.

Conclusion: CPP and H₁-H₂ respond to changes in phonation patterns. The results indicate within-person changes in each measure could indicate a change in vocal function. In particular, pressed phonation resulted in increased CPP in all speech tasks for females and in connected speech for males, suggesting that CPP increasing too much could be associated with vocal hyperfunction.