📫 Summary:
The implemented 3D articulatory model synthesizes speech acoustic for static vocal tract shapes. The model couples a lumped-element vocal fold model [1] (i.e., two-mass model) with the 3D vocal tract model to generate synthetic audio output for various vocal tract shapes. During the simulation, the model uses the 1D area functions dataset [2] (i.e., change in vocal tract cross-sectional shape from glottis to lips) to generate the 3D vocal tract contour. The synthesizer then uses the finite-difference time-domain numerical scheme [3][4] to discretize acoustic components (pressure and velocity) on a staggered othrogonal grid and computes the acoustic pressure wave propagation. The implemented model is motivated from Takemoto's 3D vocal tract model [3] and "Aerophones In Flatland" [5] research articles. It also faciliates the transfer function analysis of static vocal tracts having different cross-sectional shapes (i.e., circular, elliptical and square).
🎮 Start the simulator:
- To simulate the articulatory model, use the main.m file in the MATLAB environemnt.
- For transfer function analysis, use the plotFrequencyPhase.m file.
- To geneate audio samples, use the generateAudio.m file.
📚 References:
If you use the code for your research work, please cite the following papers -
[1] "Three-dimensional finite-difference time-domain acoustic analysis of simplified vocal tract shapes" by Mohapatra et al.
@inproceedings{mohapatra22_interspeech,
author={Debasish Mohapatra and Mario Fleischer and Victor Zappi and Peter Birkholz and Sidney Fels},
title={{Three-dimensional finite-difference time-domain acoustic analysis of simplified vocal tract shapes}},
year=2022,
booktitle={Proc. Interspeech 2022},
pages={764--768},
doi={10.21437/Interspeech.2022-10649}
}
[2] "An Extended Two-Dimensional Vocal Tract Model for Fast Acoustic Simulation of Single-Axis Symmetric Three-Dimensional Tubes" by Mohapatra et al.
@inproceedings{mohapatra19_interspeech,
author={Debasish Ray Mohapatra and Victor Zappi and Sidney Fels},
title={{An Extended Two-Dimensional Vocal Tract Model for Fast Acoustic Simulation of Single-Axis Symmetric Three-Dimensional Tubes}},
year=2019,
booktitle={Proc. Interspeech 2019},
pages={3760--3764},
doi={10.21437/Interspeech.2019-1764}
}
⛳ Future work:
- Implementation of bent vocal tract geometries having simplified cross-sections.
- Studying the effect of side branches/cavities.
- Simulation of realistic vocal tract models.
- Implementation of mouth radiation.
- Implementation of an accelerated 3D FDTD vocal tract model using modern GPUs [CUDA/OpenCL environment]
For bugs/suggestion/collaboration, please contact: debasishray@ece.ubc.ca