Objective
Clinical methods for auditory-perceptual evaluation of disordered voice quality (VQ) are highly efficient but lack the reliability and mathematical rigor needed to accurately inform and track clinical progress. Traditional psychophysical measurement methods increase reliability and rigor but are time consuming. Psychophysical scales are intuitive and allow reliable description of a change in perceived magnitude of a physical attribute (e.g., “sone” scale for loudness). Initially, we developed psychophysical scales for breathiness and for roughness. However, since dysphonic voices often co-vary along multiple dimensions, a three-dimensional (breathiness, roughness, strain) scale may be more practical and parsimonious than individual VQ scales. The current study combines data from a prior 3D matching task with new data from a 3D magnitude estimation task to develop a 3D psychophysical scale that accounts for natural covariation of these VQs.
Methods
Stimulus continua varying across three VQ dimensions were created using low-pass filtered sawtooth waveforms as the base stimulus. Adjustments were made to the signal-to-noise ratio to manipulate breathiness, amplitude modulation depth to modify roughness, and bandpass filter gain to modify strain, each in 11 steps based on the range of prior matching values. For each step, the other two VQ variables were adjusted across 9 covarying conditions. Thirteen inexperienced listeners assessed the perceived magnitudes of the three VQs using a 3D magnitude estimation task.
Results
Intra- and interrater reliability were high (ICC > 0.89). For each VQ dimension, a modified logistic equation resulted in an excellent fit (R_ > 0.9) between the mean magnitude estimates and the acoustic changes implemented to manipulate VQ. The point of inflection for each dimension was established as a standard reference unit and perceived magnitudes were rescaled relative to the reference unit. This rescaled function is proposed as a psychophysical scale for each VQ dimension.
Conclusion
High rater reliability and R2 and correspondence to earlier one-dimensional scales are indicative of the successful development of 3D-scales for voice quality. These scales provide ratio-level data and support the quantification of perceived VQ for the accurate assessment of disordered voices to be implemented in a user-friendly clinical software package.
[Work supported by NIH R01DC009029]