A quantitative comparison of linear and nonlinear loudspeaker models

This repository provides a Python + JAX implementation of loudspeaker differential-equation models and fitting procedures to reproduce the results from our paper for ICA2022.

The code was tested on Linux only, apart from `notebooks/00_data_aquisition.ipynb, which was run from Windows.

Paper:

Find it here.

Data set:

Download files from the release page.

License:

MIT -- see the file LICENSE for details.

Project overview

Repo is structured as follows:

.
├── data                           # [data set, get it from release page]
├── notebooks                      # [main notebooks and scripts]
│   ├── 00_data_aquisition.ipynb       # measurement
│   ├── 01_preprocessing.ipynb         # average and cleanup data
│   ├── 02_model_training.py           # fit ode models
│   ├── 03_model_prediction.py         # predict with fitted models
│   └── 04_analysis.ipynb              # analyze results
├── src                            # [models defs, training procedures, etc.]
├── README.rst
├── environment.yml
...

Getting started

On Linux, create a virtual environment with:

conda env create -f environment.yml

On Windows, use:

Conda env create -f environment_windows.yml

Afterwards, activate the environment:

conda activate mod_comp

Download dataset from release page:

python data/download_files.py

Run scripts or notebooks in notebooks dir. Enjoy!

Citation

If you found this codebase useful in your research, please cite:

@inproceedings{heuchelQuantComp2022,
   author       = "Heuchel, Franz M. and Agerkvist, Finn T.",
   title        = "A quantitative comparison of linear and nonlinear loudspeaker models",
   booktitle    = "Proceedings of the 24th International Congress on Acoustics",
   year         = "2022",
   pages        = "1-8",
}