ISCA Archive MAVEBA 2005
ISCA Archive MAVEBA 2005

From vocal quality measurement to perception

Rahul Shrivastav

Quantification of vocal quality of a speech signal is essential in a number of applications. However, existing measures for this purpose are often characterized by poor sensitivity and specificity to perceptual judgments. These shortcomings may have arisen because (1) these measures are often validated against “noisy” perceptual data and (2) the non-linear and multidimensional relationships between the physical signal and perceptual judgments have often been ignored. This paper describes the psychometric principles underlying quantification of subjective judgments and the use of an auditory processing model as a signal processing front-end when measuring “breathy” voice quality. Preliminary data for quantification of “roughness” is also discussed.