A 2.5-dimensional acoustic wave solver: Modeling simplified vocal tract geometries with reduced computational load

Area of Science:

• Acoustics
• Computational physics
• Speech science

Background:

• High-fidelity 3D wave solvers are accurate but computationally expensive for vocal tract modeling.
• Low-dimensional models are efficient but lack accuracy for complex geometries and higher-order modes.

Purpose of the Study:

• Introduce a lightweight 2.5D lumped solver for efficient and accurate vocal tract acoustic modeling.
• Evaluate the 2.5D solver's performance against 2D and 3D models for straight vocal tract geometries.

Main Methods:

• Developed a lumped two-dimensional (2.5D) solver for straight tracts with mid-sagittal symmetry.
• Validated the 2.5D model by comparing transfer functions and pressure distributions with 2D and 3D finite element models.
• Tested the solver on six straight vocal tract geometries of varying complexity.

Main Results:

• The 2.5D solver closely matched 3D model transfer functions up to 12 kHz (correlation > 0.8 for symmetric tracts).
• The 2.5D model significantly outperformed the 2D model for asymmetric geometries.
• Achieved over two orders of magnitude computational speed-up compared to the 3D model.

Conclusions:

• The 2.5D solver provides a favorable accuracy-efficiency trade-off for vocal tract acoustic modeling.
• It captures transverse wave propagation and higher-order modes, improving upon low-dimensional models.
• The method shows promise for real-time applications in speech synthesis and analysis.