Multiparametric Acoustic Predictors: A Comprehensive Model for Auditory Perception of Overall Severity of Vocal Deviation in Brazilian Portuguese Speakers.

Objective: To develop an acoustic model to predict the variability in the auditory perception of the overall severity (OS) of vocal deviation in male and female voices of Brazilian Portuguese speakers.

Method: In total, 300 vocal samples (sustained vowel [a] and connected speech) from adults of both sexes (235 women and 65 men) were analyzed, including 269 individuals with dysphonia and 31 individuals without dysphonia. Five speech-language pathologists carried out the auditory-perceptual judgment of the OS using an analog-visual scale of 100 mm. A script in the free Praat software was used to extract 47 acoustic measurements. Three linear regression models were developed using the stepwise technique to select the acoustic measurements capable of explaining the variability of the auditory-perceptual judgment of the OS in the general group and for male and female voices separately.

Results: The general model accounted for 66.3% of the variability in OS perception using five acoustic measures, illustrating how periodicity, harmonic structure, and spectral noise across various frequency bands interact to elucidate the clinical perception of vocal quality. For male voices, a more streamlined model employing only two acoustic measures explained 64.9% of the variability, emphasizing that regularity in glottal cycles and low-frequency noise are pivotal factors in the perceptual assessment of male vocal quality. The model for female voices achieved the highest explanatory power at 71.2% with five acoustic measures, suggesting that OS perception in women involves a complex interplay of signal periodicity, noise components across different frequency bands, and variations in fundamental frequency, highlighting the multifaceted nature of assessing female vocal quality.

Conclusions: The acoustic models developed predicted 66% of the variability in the perception of OS for voices in general, 65% of the variability in the perception of OS for male voices, and 71% of the variability in OS for female voices.