Related Experiment Video
Updated: Aug 27, 2025
09:58
Investigating the Three-dimensional Flow Separation Induced by a Model Vocal Fold Polyp
Published on: February 3, 2014
8.6K
Estimating vocal tract geometry from acoustic impedance using deep neural network.
Balamurali B T1, Saumitra Kapoor1, Jer-Ming Chen1
1Singapore University of Technology and Design, Singapore balamurali_bt@sutd.edu.sg, saumitrakapoor@gmail.com, jerming_chen@sutd.edu.sg.
JASA Express Letters
|September 26, 2022
Summary
This study uses artificial neural networks to determine vocal tract geometry from acoustic impedance. Standardizing impedance data significantly improved the accuracy of vocal tract model predictions.
Area of Science:
- Acoustics
- Biomedical Engineering
- Artificial Intelligence
Background:
- The inverse area function problem is crucial for understanding vocal tract acoustics.
- Accurate vocal tract geometry reconstruction is essential for speech synthesis and analysis.
Purpose of the Study:
- To develop a data-driven method using artificial neural networks to solve the inverse area function problem.
- To predict vocal tract geometry from acoustic impedance spectra.
Main Methods:
- Employing artificial neural networks for a data-driven approach.
- Modeling the vocal tract as a tube with nonuniform cylindrical cross-sections.
- Analyzing the acoustic impedance spectrum to infer geometry.
Main Results:
- High correlation (Pearson's ρ and Lin's ρc > 95%) between predicted and actual radii for 3- and 4-cylinder models.
- Lower correlation (ρ ~75%, ρc ~69%) for the 6-cylinder model without impedance standardization.
- Significant improvement in correlation (ρ and ρc > 90%) across all models after standardizing impedance values.
Conclusions:
- Artificial neural networks offer a viable approach for inverse area function estimation.
- Standardizing acoustic impedance data is critical for improving the accuracy of vocal tract geometry reconstruction.
- The method shows promise for applications in speech science and bioacoustics.
