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

Objective: Strain and roughness are psychoacoustic properties commonly used as parameters in clinical voice evaluations. While many studies have investigated the acoustic properties related to the perception of roughness, few studies have examined acoustic variables related to the perception of strain. However, results from these few studies have shown a relationship between acoustic measures and the perception of strain. Based on previous research, there appears to be an overlap in acoustic variables that are related to both the perception of strain and the perception of roughness. The present study seeks to disentangle a listener’s perception of strain and roughness through a paired-comparison task with multidimensional scaling analyses and to identify acoustic correlates to emergent dimensions.

Methods/Design: For each pitch, A3 and F5, fifteen synthetic female singing voices will be generated on the vowel /ɑ/ with varying spectral source slopes of 0, -6, and -12 dB/octave and an amplitude modulation frequency of 30 Hz with depths of 0, -5, -10, -15, and -20 dB to simulate levels of strain and roughness, respectively. Within each pitch, the fifteen stimuli will be combined in all possible pairs, resulting in 210 stimuli pairs total. Experienced listeners and inexperienced listeners will be asked to rate the dissimilarity of each paired stimulus using a visual analog scale. Multidimensional scaling analyses will be conducted to visualize the perceptual dimensions hypothesized to be associated with the perception of strain and roughness. Correlation analyses will be performed on the multidimensional scaling solutions using acoustic variables to identify trends that may emerge.

Results: Based on previous research showing a relationship between spectral source slope and the perception of strain and a relationship between amplitude modulation depth and the perception of roughness, it is anticipated that, in the absence of a predetermined classification scheme, multidimensional scaling solutions will represent the perceptual spaces of strain and roughness, and that acoustic measures will correlate with specific dimensions.

Conclusion: Results from this study may improve our understanding of the complexities and potential interactions of the perception of strain and roughness and help disentangle acoustic measures related to both parameters.