Bayesian Inference of Vocal Fold Material Properties from Glottal Area Waveforms Using a 2D Finite Element Model

Sensitivity of Phonation Onset Pressure to Vocal Fold Stiffness Distribution

Phonation onset is characterized by the unstable growth of vocal fold (VF) vibrations that ultimately results in self-sustained oscillation and the production of modal voice. Motivated by histological studies, much research has focused on the role of the layered structure of the vocal folds in influencing phonation onset, wherein the outer "cover" layer is relatively soft and the inner "body" layer is relatively stiff. Recent research, however, suggests that the body-cover (BC) structure over-simplifies actual stiffness distributions by neglecting important spatial variations, such as inferior-superior (IS) and anterior-posterior gradients and smooth transitions in stiffness from one histological layer to another. Herein, we explore sensitivity of phonation onset to stiffness gradients and smoothness. By assuming no a priori stiffness distribution and considering a second-order Taylor series sensitivity analysis of phonation onset pressure with respect to stiffness, we find two general smooth stiffness distributions most strongly influence onset pressure: a smooth stiffness containing aspects of BC differences and IS gradients in the cover, which plays a role in minimizing onset pressure, and uniform increases in stiffness, which raise onset pressure and frequency. While the smooth stiffness change contains aspects qualitatively similar to layered BC distributions used in computational studies, smooth transitions in stiffness result in higher sensitivity of onset pressure than discrete layering. These two general stiffness distributions also provide a simple, low-dimensional, interpretation of how complex variations in VF stiffness affect onset pressure, enabling refined exploration of the effects of stiffness distributions on phonation onset.

Synthetic, self-oscillating vocal fold models for voice production researcha)

Sound for the human voice is produced by vocal fold flow-induced vibration and involves a complex coupling between flow dynamics, tissue motion, and acoustics. Over the past three decades, synthetic, self-oscillating vocal fold models have played an increasingly important role in the study of these complex physical interactions. In particular, two types of models have been established: "membranous" vocal fold models, such as a water-filled latex tube, and "elastic solid" models, such as ultrasoft silicone formed into a vocal fold-like shape and in some cases with multiple layers of differing stiffness to mimic the human vocal fold tissue structure. In this review, the designs, capabilities, and limitations of these two types of models are presented. Considerations unique to the implementation of elastic solid models, including fabrication processes and materials, are discussed. Applications in which these models have been used to study the underlying mechanical principles that govern phonation are surveyed, and experimental techniques and configurations are reviewed. Finally, recommendations for continued development of these models for even more lifelike response and clinical relevance are summarized.

The effect of swelling on vocal fold kinematics and dynamics

Swelling in the vocal folds is caused by the local accumulation of fluid, and has been implicated as a phase in the development of phonotraumatic vocal hyperfunction and related structural pathologies, such as vocal fold nodules. It has been posited that small degrees of swelling may be protective, but large amounts may lead to a vicious cycle wherein the engorged folds lead to conditions that promote further swelling, leading to pathologies. As a first effort to explore the mechanics of vocal fold swelling and its potential role in the etiology of voice disorders, this study employs a finite-element model with swelling confined to the superficial lamina propria, which changes the volume, mass, and stiffness of the cover layer. The impacts of swelling on a number of vocal fold kinematic and damage measures, including von Mises stress, internal viscous dissipation, and collision pressure, are presented. Swelling has small but consistent effects on voice outputs, including a reduction in fundamental frequency with increasing swelling (10 Hz at 30 % swelling). Average von Mises stress decreases slightly for small degrees of swelling but increases at large magnitudes, consistent with expectations for a vicious cycle. Both viscous dissipation and collision pressure consistently increase with the magnitude of swelling. This first effort at modeling the impact of swelling on vocal fold kinematics, kinetics, and damage measures highlights the complexity with which phonotrauma can influence performance metrics. Further identification and exploration of salient candidate measures of damage and refined studies coupling swelling with local phonotrauma are expected to shed further light on the etiological pathways of phonotraumatic vocal hyperfunction.

Voice Feature Selection to Improve Performance of Machine Learning Models for Voice Production Inversion

OBJECTIVE

Estimation of physiological control parameters of the vocal system from the produced voice outcome has important applications in clinical management of voice disorders . Previously we developed a simulation-based neural network for estimation of vocal fold geometry, mechanical properties, and subglottal pressure from voice outcome features that characterize the acoustics of the produced voice. The goals of this study are to (1) explore the possibility of improving the estimation accuracy of physiological control parameters by including voice outcome features characterizing vocal fold vibration; and (2) identify voice feature sets that optimize both estimation accuracy and robustness to measurement noise.

METHODS

Feedforward neural networks are trained to solve the inversion problem of estimating the physiological control parameters of a three-dimensional body-cover vocal fold model from different sets of voice outcome features that characterize the simulated voice acoustics, glottal flow, and vocal fold vibration. A sensitivity analysis is then performed to evaluate the contribution of individual voice features to the overall performance of the neural networks in estimating the physiologic control parameters.

RESULTS AND CONCLUSIONS

While including voice outcome features characterizing vocal fold vibration increases estimation accuracy, it also reduces the network's robustness to measurement noise, due to high sensitivity of network performance to voice outcome features measuring the absolute amplitudes of the glottal flow and area waveforms, which are also difficult to measure accurately in practical applications. By excluding such glottal flow-based features and replacing glottal area-based features by their normalized counterparts, we are able to significantly improve both estimation accuracy and robustness to noise. We further show that similar estimation accuracy and robustness can be achieved with an even smaller set of voice outcome features by excluding features of small sensitivity.

Predicting 3D soft tissue dynamics from 2D imaging using physics informed neural networks

Tissue dynamics play critical roles in many physiological functions and provide important metrics for clinical diagnosis. Capturing real-time high-resolution 3D images of tissue dynamics, however, remains a challenge. This study presents a hybrid physics-informed neural network algorithm that infers 3D flow-induced tissue dynamics and other physical quantities from sparse 2D images. The algorithm combines a recurrent neural network model of soft tissue with a differentiable fluid solver, leveraging prior knowledge in solid mechanics to project the governing equation on a discrete eigen space. The algorithm uses a Long-short-term memory-based recurrent encoder-decoder connected with a fully connected neural network to capture the temporal dependence of flow-structure-interaction. The effectiveness and merit of the proposed algorithm is demonstrated on synthetic data from a canine vocal fold model and experimental data from excised pigeon syringes. The results showed that the algorithm accurately reconstructs 3D vocal dynamics, aerodynamics, and acoustics from sparse 2D vibration profiles.

Estimation of Subglottal Pressure, Vocal Fold Collision Pressure, and Intrinsic Laryngeal Muscle Activation From Neck-Surface Vibration Using a Neural Network Framework and a Voice Production Model

The ambulatory assessment of vocal function can be significantly enhanced by having access to physiologically based features that describe underlying pathophysiological mechanisms in individuals with voice disorders. This type of enhancement can improve methods for the prevention, diagnosis, and treatment of behaviorally based voice disorders. Unfortunately, the direct measurement of important vocal features such as subglottal pressure, vocal fold collision pressure, and laryngeal muscle activation is impractical in laboratory and ambulatory settings. In this study, we introduce a method to estimate these features during phonation from a neck-surface vibration signal through a framework that integrates a physiologically relevant model of voice production and machine learning tools. The signal from a neck-surface accelerometer is first processed using subglottal impedance-based inverse filtering to yield an estimate of the unsteady glottal airflow. Seven aerodynamic and acoustic features are extracted from the neck surface accelerometer and an optional microphone signal. A neural network architecture is selected to provide a mapping between the seven input features and subglottal pressure, vocal fold collision pressure, and cricothyroid and thyroarytenoid muscle activation. This non-linear mapping is trained solely with 13,000 Monte Carlo simulations of a voice production model that utilizes a symmetric triangular body-cover model of the vocal folds. The performance of the method was compared against laboratory data from synchronous recordings of oral airflow, intraoral pressure, microphone, and neck-surface vibration in 79 vocally healthy female participants uttering consecutive /pæ/ syllable strings at comfortable, loud, and soft levels. The mean absolute error and root-mean-square error for estimating the mean subglottal pressure were 191 Pa (1.95 cm H2O) and 243 Pa (2.48 cm H2O), respectively, which are comparable with previous studies but with the key advantage of not requiring subject-specific training and yielding more output measures. The validation of vocal fold collision pressure and laryngeal muscle activation was performed with synthetic values as reference. These initial results provide valuable insight for further vocal fold model refinement and constitute a proof of concept that the proposed machine learning method is a feasible option for providing physiologically relevant measures for laboratory and ambulatory assessment of vocal function.

Vocal fold dynamics in a synthetic self-oscillating model: Contact pressure and dissipated-energy dose

The energy dissipated during vocal fold (VF) contact is a predictor of phonotrauma. Difficulty measuring contact pressure has forced prior energy dissipation estimates to rely upon generalized approximations of the contact dynamics. To address this shortcoming, contact pressure was measured in a self-oscillating synthetic VF model with high spatiotemporal resolution using a hemilaryngeal configuration. The approach yields a temporal resolution of less than 0.26 ms and a spatial resolution of 0.254 mm in the inferior-superior direction. The average contact pressure was found to be 32% of the peak contact pressure, 60% higher than the ratio estimated in prior studies. It was found that 52% of the total power was dissipated due to collision. The power dissipated during contact was an order of magnitude higher than the power dissipated due to internal friction during the non-contact phase of oscillation. Both the contact pressure magnitude and dissipated power were found to be maximums at the mid anterior-posterior position, supporting the idea that collision is responsible for the formation of benign lesions, which normally appear at the middle third of the VF.

The influence of flow model selection on finite element model parameter estimation using Bayesian inference

Recently, Bayesian estimation coupled with finite element modeling has been demonstrated as a viable tool for estimating vocal fold material properties from kinematic information obtained via high-speed video recordings. In this article, the sensitivity of the parameter estimations to the employed fluid model is explored by considering Bernoulli and one-dimensional viscous fluid flow models. Simulation results indicate that prescribing an ad hoc separation location for the Bernoulli flow model can lead to large estimate biases, whereas including the separation location as an estimated parameter leads to results comparable to that of the viscous fluid flow model.

A one-dimensional flow model enhanced by machine learning for simulation of vocal fold vibration

A one-dimensional (1D) unsteady and viscous flow model that is derived from the momentum and mass conservation equations is described, and to enhance this physics-based model, a machine learning approach is used to determine the unknown modeling parameters. Specifically, an idealized larynx model is constructed and ten cases of three-dimensional (3D) fluid-structure interaction (FSI) simulations are performed. The flow data are then extracted to train the 1D flow model using a sparse identification approach for nonlinear dynamical systems. As a result of training, we obtain the analytical expressions for the entrance effect and pressure loss in the glottis, which are then incorporated in the flow model to conveniently handle different glottal shapes due to vocal fold vibration. We apply the enhanced 1D flow model in the FSI simulation of both idealized vocal fold geometries and subject-specific anatomical geometries reconstructed from the magnetic resonance imaging images of rabbits' larynges. The 1D flow model is evaluated in both of these setups and shown to have robust performance. Therefore, it provides a fast simulation tool that is superior to the previous 1D models.

Numerical and experimental investigations on vocal fold approximation in healthy and simulated unilateral vocal fold paralysis

We have developed a novel surgical/computational model for the investigation of unilateral vocal fold paralysis (UVFP) which will be used to inform future in silico approaches to improve surgical outcomes in type I thyroplasty. Healthy phonation (HP) was achieved using cricothyroid suture approximation on both sides of the larynx to generate symmetrical vocal fold closure. Following high-speed videoendoscopy (HSV) capture, sutures on the right side of the larynx were removed, partially releasing tension unilaterally and generating asymmetric vocal fold closure characteristic of UVFP (sUVFP condition). HSV revealed symmetric vibration in HP, while in sUVFP the sutured side demonstrated a higher frequency (10 - 11%). For the computational model, ex vivo magnetic resonance imaging (MRI) scans were captured at three configurations: non-approximated (NA), HP, and sUVFP. A finite-element method (FEM) model was built, in which cartilage displacements from the MRI images were used to prescribe the adduction and the vocal fold deformation was simulated before the eigenmode calculation. The results showed that the frequency comparison between the two sides were consistent with observations from HSV. This alignment between the surgical and computational models supports the future application of these methods for the investigation of treatment for UVFP.

Bayesian estimation of vocal function measures using laryngeal high-speed videoendoscopy and glottal airflow estimates: An in vivo case study

This study introduces the in vivo application of a Bayesian framework to estimate subglottal pressure, laryngeal muscle activation, and vocal fold contact pressure from calibrated transnasal high-speed videoendoscopy and oral airflow data. A subject-specific, lumped-element vocal fold model is estimated using an extended Kalman filter and two observation models involving glottal area and glottal airflow. Model-based inferences using data from a vocally healthy male individual are compared with empirical estimates of subglottal pressure and reference values for muscle activation and contact pressure in the literature, thus providing baseline error metrics for future clinical investigations.

Estimation of vocal fold physiology from voice acoustics using machine learning

The goal of this study is to estimate vocal fold geometry, stiffness, position, and subglottal pressure from voice acoustics, toward clinical and other voice technology applications. Unlike previous voice inversion research that often uses lumped-element models of phonation, this study explores the feasibility of voice inversion using data generated from a three-dimensional voice production model. Neural networks are trained to estimate vocal fold properties and subglottal pressure from voice features extracted from the simulation data. Results show reasonably good estimation accuracy, particularly for vocal fold properties with a consistent global effect on voice production, and reasonable agreement with excised human larynx experiment.

Risk Quantification and Analysis of Coupled Factors Based on the DEMATEL Model and a Bayesian Network

With the developing of high integrations in large scale systems, such as aircraft and other industrial systems, there are new challenges in safety analysis due to the complexity of the mission process and the more complicated coupling characteristic of multi-factors. Aiming at the evaluation of coupled factors as well as the risk of the mission, this paper proposes a combined technology based on the Decision Making Trial and Evaluation Laboratory (DEMATEL) model and the Bayesian network (BN). After identifying and classifying the risk factors from the perspectives of humans, machines, the environment, and management, the DEMATEL technique is adopted to assess their direct and/or indirect coupling relationships to determine the importance and causality of each factor; moreover, the relationship matrix in the DEMATEL model is used to generate the BN model, including its parameterization. The inverse reasoning theory is then implemented to derive the probability, and the risk of the coupled factors is evaluated by an assessment model integrating the probability and severity. Furthermore, the key risk factors are identified based on the risk radar diagram and the Pareto rule to support the preventive measurements. Finally, an application of the take-off process of aircraft is provided to demonstrate the proposed method.

Toward Development of a Vocal Fold Contact Pressure Probe: Bench-Top Validation of a Dual-Sensor Probe Using Excised Human Larynx Models

A critical element in understanding voice production mechanisms is the characterization of vocal fold collision, which is widely considered a primary etiological factor in the development of common phonotraumatic lesions such as nodules and polyps. This paper describes the development of a transoral, dual-sensor intraglottal/subglottal pressure probe for the simultaneous measurement of vocal fold collision and subglottal pressures during phonation using two miniature sensors positioned 7.6 mm apart at the distal end of a rigid cannula. Proof-of-concept testing was performed using excised whole-mount and hemilarynx human tissue aerodynamically driven into self-sustained oscillation, with systematic variation of the superior-inferior positioning of the vocal fold collision sensor. In the hemilarynx experiment, signals from the pressure sensors were synchronized with an acoustic microphone, a tracheal-surface accelerometer, and two high-speed video cameras recording at 4000 frames per second for top-down and en face imaging of the superior and medial vocal fold surfaces, respectively. As expected, the intraglottal pressure signal exhibited an impulse-like peak when vocal fold contact occurred, followed by a broader peak associated with intraglottal pressure build-up during the de-contacting phase. As subglottal pressure was increased, the peak amplitude of the collision pressure increased and typically reached a value below that of the average subglottal pressure. Results provide important baseline vocal fold collision pressure data with which computational models of voice production can be developed and in vivo measurements can be referenced.