and RL Erickson RA Cleveland J. Hillebrand

Hillenbrand et al. attempt to evaluate the reliability of several acoustic measures for predicting perceptual breathiness, finding cepstrum (a Fourier transform of a spectral slice) to be the most effective. Acoustic measures are made of periodicity (cepstrum), H1 amplitude,and spectral tilt. Among the use of this information for understanding pathological breathiness, phonological breathiness, and VQ changes in aging, the authors note the possibility of gender differences in breathiness levels (citing i.a. McKay 1987). Several later writers (e.g. Esposito 2006 above) have identified cepstral peak prominence as the most reliable measure of perceptual voice quality change for English listeners.


Speakers were trained to produce moderately and very breathy vowels and made familiar with examples, but the reliability of such a method for interspeaker similarity of tokens is perhaps open to question. Additionally, male “very breathy” tokens were rated as breathier than female “very breathy” tokens, which the authors suggest may be a listener compensation for anticipated lower levels of breathiness in men (leading to tokens with similar levels of breathiness being perceived as more abnormal and consequently “breathier”). While I have limited ability to assess the reliability of Hillebrand et al.’s results due to limited understanding of the statistical methods involved, I chose to include this paper in my reading list to get a stronger sense of what is meant by cepstral peak prominence and its touted reliability as a VQ measure in English. The difficulty and technical knowledge required to implement this method makes me more comfortable with a simple H1-H2 measure as implemented by e.g. Szakay (2010, below). Article appeared in Journal of Speech and Hearing Research 37/4.… (lisätietoja)