Session Q4: Phonation and Speech

A comparison of self-oscillating phonation models

This talk presents a comparison of self-oscillating models of phonation. The goal is to assess how well synthetic rubber vocal folds reproduce the gross behavior of phonation. Data from molded rubber folds and a variety of excised mammalian larynges were collected from the literature and from the authors' physical model. Gross trends are discussed and a simple scaling is presented that appears to collapse these data. Finally, comparisons between molded rubber folds and excised larynges are highlighted.   

Correlation of phonatory behavior with vocal fold structure, observed in a physical model

The effect of vocal fold shape and internal structure on phonation was studied experimentally using a physical model of the human airway. Model folds used a ``M5'' or a swept ellipse coronal cross-section shape. Models were molded in either 2 or three layers. Two-layer models included a more stiff ``body'' layer and a much softer ``cover'' layer, while the 3-layer models also incorporated an additional, thin, ``ligament/conus'' layer stiffer than the body layer. The elliptical section models were all molded in 3 such layers. Measurements of transglottal pressure, volume flow, mouth sound pressure, and high-speed imaging of vocal fold vibration were performed. These show that models with the ``ligament'' layer experienced much attenuated vertical deformation, that glottal closure was more likely, and that phonation was much easier to initiate. These findings suggest that the combination of the vocal ligament and the \textit{conus elasticus} stabilize the vocal fold for efficient phonation by limiting vertical deformation, while allowing transverse deformations to occur.   

Modeling Vocal Fold Intravascular Flow using Synthetic Replicas

Vocal fold vibration that is induced by air flowing from the lungs is believed to decrease blood flow through the vocal folds. This is important due to the critical role of blood flow in maintaining tissue health. However, the precise mechanical relationships between vocal fold vibration and blood perfusion remain understudied. A platform for studying liquid perfusion in a synthetic, life-size, self-oscillating vocal fold replica has recently been developed. The replicas are fabricated using molded silicone with material properties comparable to those of human vocal fold tissues and that include embedded microchannels through which liquid is perfused. The replicas are mounted on an air flow supply tube to initiate flow-induced vibration. A liquid reservoir is attached to the microchannel to cause liquid to perfuse through replica in the anterior-posterior direction. As replica vibration is initiated and amplitude increases, perfusion flow rate decreases. In this presentation, the replica design will be presented, along with data quantifying the relationships between parameters such as replica vibration amplitude, stiffness, microchannel diameter, and perfusion flow rate.   

Simultaneous temporally resolved DPIV and pressure measurements of symmetric oscillations in a scaled-up vocal fold model

The objective of this study is to apply experimental data to theoretical framework of Krane (2013) in which the principal aeroacoustic source is expressed in terms of vocal fold drag, glottal jet dynamic head, and glottal exit volume flow, reconciling formal theoretical aeroacoustic descriptions of phonation with more traditional lumped-element descriptions. These quantities appear in the integral equations of motion for phonatory flow. In this way time resolved velocity field measurements can be used to compute time-resolved estimates of the relevant terms in the integral equations of motion, including phonation aeroacoustic source strength. A simplified 10x scale vocal fold model from Krane, \textit{et al}. (2007) was used to examine symmetric, $i.e.$ `healthy', oscillatory motion of the vocal folds. By using water as the working fluid, very high spatial and temporal resolution was achieved. Temporal variation of transglottal pressure was simultaneously measured with flow on the vocal fold model mid-height. Experiments were dynamically scaled to examine a range of frequencies corresponding to male and female voice. The simultaneity of the pressure and flow provides new insights into the aeroacoustics associated with vocal fold oscillations.   

Examining diseased states in a scaled-up vocal fold model using simultaneous temporally resolved DPIV and pressure measurements

This study builds on the parallel presentation of Ringenberg, \textit{et al}. (APS-DFD 2017) involving simultaneous, temporally and spatially resolved flow and pressure measurements in a scaled-up vocal fold model. In this talk, data from experiments replicating characteristics of diseased vocal folds are presented. This begins with vocal folds that do not fully close and continues with asymmetric oscillations. Data are compared to symmetric, $i.e.$ `healthy', oscillatory motions presented in the companion talk. Having pressure and flow data for individual as well as phase averaged oscillations for these diseased cases highlights the potential for aeroacoustic analysis in this complex system.   

Effect of pneumotach on measurement of vocal function

Aerodynamic and acoustic measurements of vocal function were performed in a physical model of the human airway with and without a pneumotach (Rothenberg mask), used by clinicians to measure vocal volume flow. The purpose of these experiments was to assess whether the device alters acoustic and aerodynamic conditions sufficiently to change phonation behavior. The airway model, which mimics acoustic behavior of an adult human airway from trachea to mouth, consists of a 31.5cm long straight duct with a 2.54cm square cross section. Model vocal folds comprised of molded silicone rubber were set into vibration by introducing airflow from a compressed air source. Measurements included transglottal pressure difference, mean volume flow, vocal fold vibratory motion, and sound pressure measured at the mouth. The experiments show that while the pneumotach imparted measurable aerodynamic and acoustic loads on the system, measurement of mean glottal resistance was not affected. Acoustic pressure levels were attenuated, however, suggesting clinical acoustic measurements of vocal function need correction when performed in conjunction with a pneumotach   

Aerosol emission during human speech

We show that the rate of aerosol particle emission during healthy human speech is strongly correlated with the loudness (amplitude) of vocalization. Emission rates range from approximately 1 to 50 particles per second for quiet to loud amplitudes, regardless of language spoken (English, Spanish, Mandarin, or Arabic). Intriguingly, a small fraction of individuals behave as ``super emitters,'' consistently emitting an order of magnitude more aerosol particles than their peers. We interpret the results in terms of the eggressive flowrate during vocalization, which is known to vary significantly for different types of vocalization and for different individuals. The results suggest that individual speech patterns could affect the probability of airborne disease transmission. The results also provide a possible explanation for the existence of ``super spreaders'' who transmit pathogens much more readily than average and who play a key role in the spread of epidemics.   

Power Flow in Phonation

The control volume analysis of power flow during sustained phonation is performed using results of a fully-coupled aeroelastic-aeroacoustic simulation. The control volumes consist of the laryngeal region, and the larynx and the vocal tract. Two cases are considered: an effectively infinite length vocal tract, where sound produced in the larynx radiates away and is not reflected back, and a constant-area vocal tract of normal adult human dimensions, in which phonatory sound resonates before radiating from the mouth opening. In both cases the lungs are modeled to absorb all incident sound, while providing a constant volume flow toward the larynx. Control of the acoustic boundary conditions is accomplished using perfectly matched- layers, and flow from the lungs is provided by a source distribution near the entrance to the trachea region. For both cases the power flow for the larynx and larynx plus vocal tract control volumes are computed using the integral form of the mechanical energy equation, expanded to consider power exchanges between slightly compressible flow in the larynx and the acoustic fields in the vocal tract and trachea.   

Performance of a reduced-order FSI model for flow-induced vocal fold vibration

Vocal fold vibration during speech production involves a three-dimensional unsteady glottal jet flow and three-dimensional nonlinear tissue mechanics. A full 3D fluid-structure interaction (FSI) model is computationally expensive even though it provides most accurate information about the system. On the other hand, an efficient reduced-order FSI model is useful for fast simulation and analysis of the vocal fold dynamics, which can be applied in procedures such as optimization and parameter estimation. In this work, we study performance of a reduced-order model as compared with the corresponding full 3D model in terms of its accuracy in predicting the vibration frequency and deformation mode. In the reduced-order model, we use a 1D flow model coupled with a 3D tissue model that is the same as in the full 3D model. Two different hyperelastic tissue behaviors are assumed. In addition, the vocal fold thickness and subglottal pressure are varied for systematic comparison. The result shows that the reduced-order model provides consistent predictions as the full 3D model across different tissue material assumptions and subglottal pressures. However, the vocal fold thickness has most effect on the model accuracy, especially when the vocal fold is thin.   

Direct numerical simulation of human phonation

The generation and propagation of the human voice in three-dimensions is studied using direct numerical simulation. A full body domain is employed for the purpose of directly computing the sound in the region past the speaker's mouth. The air in the vocal tract is modeled as a compressible and viscous fluid interacting with the elastic vocal folds. The vocal fold tissue material properties are multi-layered, with varying stiffness, and a linear elastic transversely isotropic model is utilized and implemented in a quadratic finite element code. The fluid-solid domains are coupled through a boundary-fitted interface and utilize a Poisson equation-based mesh deformation method. A kinematic constraint based on a specified minimum gap between the vocal folds is applied to prevent collision during glottal closure. Both near VF flow dynamics and far-field acoustics have been studied. A comparison is drawn to current two-dimensional simulations as well as to data from the literature. Near field vocal fold dynamics and glottal flow results are studied and in good agreement with previous three-dimensional phonation studies. Far-field acoustic characteristics, when compared to their two-dimensional counterpart, are shown to be sensitive to the dimensionality.   

Flow-structure interaction simulation of voice production in a canine larynx.

Experimental measurements conducted on a hemi-larynx canine vocal fold showed that negative pressures formed in the glottis near the superior surface of the vocal fold in the closing phase even without a supra-glottal vocal tract. It was hypothesized that such negative pressures were due to intraglottal vortices caused by flow separation in a divergent vocal tract during vocal fold closing phase. This work aims to test this hypothesis from the numerical aspect. Flow-structure interaction simulations are performed in realistic canine laryngeal shapes. In the simulations, a sharp interface immersed boundary method based incompressible flow solver is utilized to model the air flow; a finite element based solid mechanics solver is utilized to model the vocal fold vibration. The geometric structure of the vocal fold and vocal tract are based on MRI scans of a mongrel canine. The vocal fold tissue is modeled as transversely isotropic nonlinear materials with a vertical stiffness gradient. Numerical indentation is first performed and compared with the experiment data to obtain the material properties. Simulation setup about the inlet and outlet pressure follows the setup in the experiment. Simulation results including the fundamental frequency, air flow rate, the divergent angle will be compared with the experimental data, providing the validation of the simulation approach. The relationship between flow separation, intra-glottal vortices, divergent angle and flow rate will be comprehensively analyzed.   

Physics-based analysis and control of human snoring

In order to advance the understanding of biological fluid dynamics and its effects on the acoustics of human snoring, the study pursued a physics-based computational approach. From human magnetic resonance image (MRI) scans, the researchers were able to develop both anatomically and dynamically accurate airway-uvula models. With airways defined as rigid, and the uvula defined as flexible, computational models were created with various pharynx thickness and geometries. In order to determine vortex shedding with prescribed uvula movement, the uvula fluctuation was categorized by its specific parameters: magnitude, frequency, and phase lag. Uvula vibration modes were based on one oscillation, or one harmonic frequency, and pressure probes were located in seven different positions throughout the airway-uvula model. By taking fast Fourier transforms (FFT) from the pressure probe data, it was seen that four harmonics were created throughout the simulation within one oscillation of uvula movement. Of the four harmonics, there were two pressure probes which maintained high amplitudes and led the researcher to believe that different vortices formed with different snoring frequencies.