Objective Acoustic Quantification of Perceived Voice Tremor Severity

Abstract

Purpose This study compared auditory-perceptual measures of presence/absence and severity of vocal tremor to acoustic markers of vocal tremor. The validity (both concurrent and diagnostic) of various acoustic markers of vocal tremor was also assessed. Method Fifty-six midvowel sustained [a:] recordings were selected to yield a representative convenience sample of vocal tremor. After training with 10 synthesized samples, 4 female audiologists rated these samples on "voice tremor severity" on a continuous 10-cm scale. Afterward, 15 randomly selected recordings were presented a 2nd time for intrarater reliability assessment. Customized audio signal processing in Praat yielded 12 acoustic measures of rate, extent and perturbation of fundamental frequency (f 0), and intensity level (IL) modulation. Enter-type multiple linear regression analysis was applied to weight and combine these acoustic variables into an acoustic model of vocal tremor severity. Results After removing the vocal tremor severity ratings of 1 of the audiologists because of insufficient intra- and interrater reliability, mean single-measures consistency-type intraclass correlation coefficients equaled .83 within raters and .72 between raters. Correlation between mean ratings and the 12 acoustic markers ranged from .76 for median extent of f 0 modulation to .11 for rate of IL modulation. Correlation between mean ratings and the acoustic model was .89. Analysis of this model's receiver operating characteristics yielded an area under receiver operating characteristic curve of .93, denoting sensitivity of .87 and specificity of .91. Conclusions This study demonstrated that auditory-perceptual ratings of vocal tremor severity are guided primarily by f 0 and IL modulation extent, less by modulation perturbation, and least by modulation rate. The acoustic model covering all these modulation properties yielded acceptable results in terms of both concurrent and diagnostic validity. However, external cross-validation of this model is warranted before applying it in clinical voice/speech assessment.