Objective
Auditory-perceptual evaluation of voice quality is often considered the gold standard in clinical voice assessment. However, the subjective nature of arbitrary assignment of numbers using ordinal or interval scales can limit reliability and precision for quantifying changes in response to disorder progression or treatment. To address these limitations, matching tasks with synthetic comparison sounds have been investigated to enhance auditory-perceptual evaluation methods. Building upon the success of single-variable matching tasks developed and validated in previous studies to measure individual voice quality dimensions (including breathiness, roughness, and strain), this study introduces a novel three-dimensional matching (3DMA) task for simultaneously measuring the three quality dimensions and examines the concurrent validity of the task.
Methods
Recordings of 26 natural /ɑ/ phonations were selected to include a wide variation across the three quality dimensions. The base comparison sound consisted of a low-pass filtered sawtooth waveform (fo = fc = 151Hz). To match the quality of the voice samples, participants concurrently adjusted three variables of the comparison sound: the signal-to-noise ratio for breathiness, the depth of amplitude modulation for roughness, and the gain of a bandpass filter for strain. The same participants completed a three-dimensional magnitude estimation (3DME) task, in which they assigned magnitudes (1-1000) to the same three dimensions. Computational measures associated with voice quality were obtained for the same voice samples using bio-inspired auditory models associated with the three voice quality dimensions.
Results
Both the 3DMA and 3DME tasks exhibited good reliability. There was a significant correlation (r > 0.9) between the obtained matching values and magnitude estimates for each voice quality dimension. There also was a significant correlation (r = 0.42 to 0.82) between the obtained matching values and the dimension-specific computational measures for each voice quality dimension.
Conclusions
The relationships measured and resulting correlations among the 3DMA, 3DME, and computational values strongly support the concurrent validity of the 3DMA task for the evaluation of voice quality.
[Work supported by NIH R01DC009029]
|