Correlations Between Measures of Cepstral Peak Prominence, Long-Term Average Spectrum, and Perceptual Ratings of Singing Voice Quality
Authors: Deirdre D. Michael, Ph.D., Susan Buesgens, M.A., Jennifer Swanson, M.M., Katherine Lindsay, B.A.
Introduction: An ongoing concern in the scientific study of voice is the quest for objective measures of voice quality that correlate with the “gold standard” of reliable perceptual ratings. Acoustic measures have been used with varying degrees of success to quantify voice quality for dysphonic voices, but they have been used less often and less successfully to quantify singing voice quality. This may be due both to the lack of acoustic measures and the difficulty obtaining reliable ratings of singing voice quality.
The acoustic measure of Cepstral Peak Prominence has been shown in numerous studies to correlate with measures of dysphonia, most especially the perceptual measure of “overall grade of dysphonia” in samples of voice ranging from normal to the most severely dysphonic. This study examines the correlations of CPP with ratings of singing voice quality, ranging from normal to most supranormal.
The acoustic measure of Long-term Average Spectrum (LTAS) has been used to measure singing voice quality, and has been moderately successful in correlating with ratings of dysphonic voice quality. This study compares LTAS to CPP in quantifying singing voice quality.
In studies of dysphonic voice quality, speech-language pathologists have shown high reliability in rating qualities of voice that are related to laryngeal function. This study examines whether singing teachers can achieve the same level of reliability for ratings of singing voice quality, when using similar procedures.
Methods: Recordings of a vocalise and a passage from a song were recorded from
singers ranging in proficiency from beginners to elite performers. These recordings
were subjected to CPP and LTAS analysis. The samples were also rated for degree
of overall quality of voice, breathiness, roughness, strain, and pitch stability
by singing teachers. Pearson correlation coefficients were used to compare CPP
and LTAS measures to the perceptual ratings.
Results: Results from a pilot study using 34 samples of singing rated by four singing teachers show that after a training procedure, the teachers’ ratings were highly and significantly correlated, so that a single average rating could be used to correlate with the CPP and LTAS ratings. CPP measures correlated with ratings of breathiness, but generally not with the other perceptual ratings. In general, LTAS measures correlated with ratings of breathiness and overall quality, and occasionally with measures of roughness and pitch stability.
Discussion: In this study, CPP was useful as a measure of breathiness, as opposed
to studies of dysphonia in which it has correlated better with ratings of overall
grade of dysphonia. Measures of LTAS were also most useful in measuring breathiness.
The possible reasons for this are explored, as are the possible uses for these
two measures in quantifying singing voice quality.
More importantly, in this study, singing teachers were able to differentiate
a continuum of singers along parameters that are associated with laryngeal function
and more typically used in ratings of dysphonia. It is possible that those measures
can serve as a point of departure in the ongoing quest for objective and reliable
measures of singing voice quality.
Deirdre D. Michael, Ph.D.
Lions Voice Clinic, University of Minnesota
Department of Otolaryngology, Box 478 MMC
420 Delaware Street SE
Minneapolis, MN 55455
Phone: 612-626-0925
Fax: 612-625-2101
e-mail: micha008@umn.edu