Accompanying code to the paper Room transfer function reconstruction using complex-valued neural networks Networks [1].

• Dependencies
• Data Generation
• Network Training
• Results Computation

• Python, it has been tested with version 3.9.17
• Numpy, scikit-image, scikit-learn,argparse, tqdm, matplotlib
• Pytorch 2.0.1+cu118
• complexPyTorch

Data generation code is contained in the folder create_dataset and is partially taken from Lluis et al.[2]

The training of the network uses the parameters contained in config/config.jon and can be run using main.py, which takes the following arguments:

• --config: String, path to the configuration file
• --best_model_path: String, name of the best-performing model file

By running test.py it is possible to compute the results contained in the paper. Specifically, the script computes the Normalized Mean Squared Error (NMSE) using the parameters and network model indicated by the selected configuration. The script takes the following arguments:

• --config: String, path to the configuration file
• --best_model_path: String, name of the best-performing model file

[1] Ronchini F., Comanducci L., Pezzoli M., Antonacci F. & Sarti A., Room transfer function reconstruction using complex-valued neural networks Networks, submitted to ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

[2] Lluis, F., Martinez-Nuevo, P., Bo Møller, M., & Ewan Shepstone, S. (2020). Sound field reconstruction in rooms: Inpainting meets super-resolution. The Journal of the Acoustical Society of America, 148(2), 649-659.