Related Experiment Video
Updated: Jul 1, 2025
10:16
Synthetic, Multi-Layer, Self-Oscillating Vocal Fold Model Fabrication
Published on: December 2, 2011
14.0K
Neural network-based estimation of biomechanical vocal fold parameters
Jonas Donhauser1, Bogac Tur1, Michael Döllinger1
1Division of Phoniatrics and Pediatric Audiology, Department of Otorhinolaryngology, Head and Neck Surgery, University Hospital Erlangen, Friedrich-Alexander University Erlangen-Nürnberg, Erlangen, Germany.
Frontiers in Physiology
|March 7, 2024
Summary
This study introduces a novel neural network model to accurately estimate vocal fold biomechanical properties from high-speed video data. This approach offers a faster, more comprehensive analysis of laryngeal dynamics than previous methods.
Area of Science:
- Biomechanics
- Computational modeling
- Laryngeal dynamics
Background:
- Vocal fold (VF) vibrations drive phonation, but direct measurement of biomechanical factors is limited with high-speed video (HSV) endoscopy.
- Physically based numerical models offer insights but require computationally intensive inverse problem solving to estimate biomechanical properties.
Purpose of the Study:
- To develop a convolutional recurrent neural network (CRNN) as a surrogate for the biomechanical inverse problem in vocal fold modeling.
- To enable fast and explicit computation of biomechanical parameters from HSV data.
Main Methods:
- A CRNN was trained on a physiological-based six-mass model (6 MM) of vocal fold dynamics.
- The network was validated by predicting subglottal pressure using 288 ex vivo porcine HSV recordings.
- The CRNN was also trained to predict VF mass and stiffness on synthetic data.
Main Results:
- The CRNN achieved a mean absolute error of 133 Pa (13.9%) in subglottal pressure prediction with 76.6% correlation on experimental data.
- A re-estimated fundamental frequency MAE of 15.9 Hz (9.9%) was achieved.
- Subglottal pressure was identified as the most learnable parameter during network training.
Conclusions:
- The CRNN serves as an effective surrogate for the entire biomechanical inverse problem, enabling explicit computation of fitted models.
- This approach advances the estimation of laryngeal kinematics and biomechanical model fitting for vocal fold dynamics.
