Modeling of Breathy Voice Quality Using Pitch-strength Estimates

Abstract

Background The characteristic voice quality of a speaker conveys important linguistic, paralinguistic and vocal health related information. Pitch strength refers to the salience of pitch sensation in a sound and was recently reported to be strongly correlated with the magnitude of perceived breathiness based on a small number of voice stimuli. Objective The current study examined the relationship between perceptual judgments of breathiness and computational estimates of pitch strength based on the Aud-SWIPE (P-NP) algorithm for a large number of voice stimuli (330 synthetic and 57 natural). Methods & Results Similar to the earlier study, the current results confirm a strong relationship between estimated pitch strength and listener judgments of breathiness such that low pitch strength values are associated with voices that have high perceived breathiness. Based on this result, a model was developed for the perception of breathy voice quality using a pitch strength estimator. Regression functions derived between the pitch strength estimates and perceptual judgments of breathiness obtained from matching task revealed a linear relationship for a subset of the natural stimuli. We then used this function to obtain predicted breathiness values for the synthetic and remaining natural stimuli. Conclusions Predicted breathiness values from our model were highly correlated with the perceptual data for both types of stimuli. Systematic differences between the breathiness of natural and synthetic stimuli are discussed.

Categorization in the Perception of Breathy Voice Quality and Its Relation to Voice Production in Healthy Speakers

“…We were also concerned that the individual differences in the degree of nasality might affect the measure instead of breathiness, so we decided not to include H1-H2 in our study. 2 An acoustic estimate of pitch strength is also known to capture signal periodicity and predict perceived breathiness (Eddins, Anand, Camacho, & Shrivastav, 2016). Pitch strength refers to the degree to which listeners can perceive pitch in a sound, and this perceptual measure has also shown a strong correlation with perceived breathiness (Shrivastav, Eddins, & Anand, 2012); however, pitch strength is not commonly used as an acoustic correlate of breathiness and was not included in this study.…”

The Perception of Breathiness in the Voices of Pediatric Speakers

“…To overcome the requirement of synthetic speech, we have also considered an index known as pitch strength to evaluate dysphonia in natural and synthetic speech (13,46). Borrowing from studies of the tonality conveyed by non-speech stimuli (47), samples from dysphonic voices were evaluated in a series of perceptual studies and modeled using computational methods for pitch strength (i.e., pitch salience from weak to strong).…”

“…Borrowing from studies of the tonality conveyed by non-speech stimuli (47), samples from dysphonic voices were evaluated in a series of perceptual studies and modeled using computational methods for pitch strength (i.e., pitch salience from weak to strong). Shrivastav et al (2012) reported a strong negative correlation (r = −0.989) between pitch strength judgments and loudness ratio, while Eddins et al (2016) demonstrated that computational pitch strength estimates were strongly and negatively correlated with perceived breathiness. These studies show that pitch strength or pitch salience is low for breathy voices and pitch strength increases as the vocal breathiness decreases.…”

“…As a first step, we have developed a perceptual experiment using a set of pediatric voices that were analyzed and resynthesized to support investigation of the two bio-inspired algorithms described above (loudness ratio and pitch strength, along with cepstral peak). These objective measures were selected because these have been shown to be highly predictive of breathiness in adult voices (45,46,48). The fidelity of the predictions of voice quality perception for pediatric voices from the current experiment were compared to analogous predictions of voice quality perception for adult voices from previous studies.…”

“…All children were between the ages of 3.0 and 4.0. Children's voices were chosen that had a higher average f 0 than those used in similar experiments evaluating breathiness in adult voices (46,52). The highest average f 0 value in the previous experiments was 219.4 Hz for a female speaker.…”