Generative models for sound field reconstruction

This repository contains the code for the paper Generative models for sound field reconstruction published in The Journal of the Acoustical Society of America (2023).

Contents

.
├── AUTHORS.md
├── LICENSE
├── README.md
├── bin
├── config
├── data
│   ├── Inference Data
│   ├── PlaneWaveData
│   └── WaveGAN Data
├── docs
└── src
    ├── models
    │   ├── CSGM
    │   ├── HiFiGAN
    │   ├── SEGAN
    │   └── WaveGAN
    ├── tools
    └── visualization

Usage

To create a conda environment with the required dependencies, run the following command in your terminal:

conda env create -f environment.yml

To train the models, you must first generate the synthetic data. This can be done by running the following command in the terminal:

python ./data/PlaneWaveData/PlaneWaveArrays.py --lsf_index $n --init_n_mics 100 --radius 0.5

where $n is the index which corresponds to [0 - N_max] sound fields. Each of these are a random wave field impinging on a spherical microphone array with the number of elements set by --init_n_mics and radius set by --radius.

For the CSGM model, one must first train the WaveGAN model. The data for the WaveGAN model is obtained by running the terminal command:

python data/WaveGAN Data/WaveGAN_Datasets.py.

To train the WaveGAN for the CSGM optimisation run the following command:

python ./models/WaveGAN/train_wavegan.py --config_file './config/wavegan_config.yaml'

where the --config_file argument is the path to the config file.

To train the HiFiGAN model run the following command:

python /models/HiFiGAN/train.py --config_file '/config/HiFiGAN_config.yaml'

To train the SEGAN model run the following command:

python /models/SEGAN/train.py --config_file '/config/SEGAN_config.yaml'

An example of each network's performance is given in the notebook ./src/visualization/bandwidth_extension_example.ipynb. To run this notebook, first run the script ./src/visualization/download_example_data.py to download the necessary files.

Citation

If you use this code, please cite the paper as:

author = {Fernandez-Grande,Efren  and Karakonstantis,Xenofon  and Caviedes-Nozal,Diego  and Gerstoft,Peter },
title = {Generative models for sound field reconstruction},
journal = {The Journal of the Acoustical Society of America},
volume = {153},
number = {2},
pages = {1179-1190},
year = {2023},
doi = {10.1121/10.0016896}}