Skip to content

ahogg/HRTF-upsampling-with-a-generative-adversarial-network-using-a-gnomonic-equiangular-projection

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

24 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

HRTF upsampling with a generative adversarial network using a gnomonic equiangular projection

⚠️ The latest code that was used for the LAP challenge (sonicom.eu/lap-challenge) is available here: https://github.com/ahogg/Exploring-the-impact-of-transfer-learning-on-GAN-based-HRTF-upsampling/tree/single_node_upsampling

A. O. T. Hogg, M. Jenkins, H. Liu, I. Squires, S. J. Cooper and L. Picinali: HRTF upsampling with a generative adversarial network using a gnomonic equiangular projection. In: Proc. IEEE/ACM Transactions on Audio Speech and Language Processing, vol. 32, pp. 2085-2099, 2024.

First, run:

Note: generate_projection only needs to be run once per dataset.

main.py generate_projection --hpc False --tag ari-upscale-4

to find the barycentric coordinates for each point in the cubed sphere. The --tag flag specifies the folder location of the results, and the --hpc flag changes the paths depending on whether the code is being executed remotely or locally (see the config.py).

Next run:

main.py preprocess --hpc False --tag ari-upscale-4

which interpolates data to find the HRIRs on the cubed sphere. It then obtains the HRTFs using the FFT and splits the data into train and validation sets.

Then to train the GAN on the cubed sphere data run:

main.py train --hpc False --tag ari-upscale-4

Note: The training parameters can be modified in the config.py file.

To test the GAN on the valuation set run:

main.py test --hpc False --tag ari-upscale-4

To compare the performance against the barycentric baseline and the HRTF selection baseline run:

main.py barycentric_baseline --hpc False 

and

main.py hrtf_selection_baseline --hpc False

respectively.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published