Multidimensional Perceptual Spaces of Strain and Roughness in Synthetic Female Singing Voices

Area of Science:

• Speech and Hearing Sciences
• Acoustic Phonetics
• Psychoacoustics

Background:

• Vocal strain and roughness are common, co-occurring perceptual qualities in dysphonic voices.
• Clinical voice assessments frequently evaluate these psychoacoustic percepts.
• Investigating the perceptual separability of strain and roughness is crucial for accurate voice disorder diagnosis.

Purpose of the Study:

• To determine if inexperienced and experienced listeners can differentiate between vocal strain and roughness.
• To identify acoustic correlates of the emergent perceptual dimensions of strain and roughness.
• To validate the clinical utility of perceptual voice assessments.

Main Methods:

• Synthetic female singing voices were created with controlled variations in spectral slope (strain) and amplitude modulation (roughness).
• Listeners rated the dissimilarity of paired voice stimuli using a visual-analog scale.
• Multidimensional scaling (MDS) analyses visualized perceptual dimensions, and hierarchical linear regression identified acoustic predictors.

Main Results:

• MDS analyses revealed distinct dimensions corresponding to strain and roughness perception across all listeners.
• Significant acoustic predictors were identified for both the strain and roughness dimensions.
• The acoustic correlates for roughness perception may be pitch-dependent.

Conclusions:

• The perception of vocal strain is perceptually separable from the perception of vocal roughness.
• Objective acoustic indices for strain and roughness have been identified, supporting clinical voice assessment.
• Findings validate the use of perceptual evaluations in clinical voice analysis and offer potential for objective measurement.