Psychometric measurement of vocal strain using a matching task with sustained vowel stimuli

Abstract

Psychometric measurement of voice quality is important for quantitative assessment of dysphonia. Previous work established the efficacy of a single-variable matching task to index vocal breathiness and roughness using appropriately designed comparison stimuli, providing a context-independent perceptual task that improves the validity and accuracy of perceptual data compared to magnitude estimation or rating tasks. Physiologically, vocal strain is characterized by excessive vocal fold adduction during phonation, resulting in a glottal pulse with a low duty cycle. The narrowing of the glottal pulse causes the excitation spectrum roll-off to decrease, resulting in a flattening of the spectrum and significant changes in spectral skewness and kurtosis [Moore et al., 1997]. Analogous to comparison stimuli for breathy and rough measurement, a set of noisy sawtooth waveforms was created using modified step filters that varied in spectral slope 1 dB/octave steps, producing a wide range of perceived strain in the comparison stimuli. The comparison stimuli were used in a listening experiment to measure the perceived strain of a set of 15 samples of /a/ selected from the Kay-PENTAX Disordered Voice Database using stratified sampling. The results validate the strain comparison stimuli and extend the matching task paradigm to the three primary dimensions of voice quality perception.