Deadline extension and updates due to various teething issues

61 views

Skip to first unread message

Aidan Hogg

unread,

Jun 20, 2024, 4:35:33 AMJun 20

to SONICOM LAP Challenge

Hi LAPers,

We would like to announce that the deadline for the LAP challenge has now been extended to the 1st of July, 2024.

We would also like to mention that the submission forms are now live. To submit your solutions for the LAP Challenge tasks, please fill out the forms below:

This deadline extension reflects the ongoing updates/changes we have had to make due to various teething issues we have faced. In light of this, we would like to announce a few more minor tweaks based on the feedback we have received.

Please make sure that you pull the latest version of the code, which reflects the following updates (Task 1: https://github.com/Audio-Experience-Design/LAPChallenge and Task 2: the ‘Spatial Audio Metrics’ library v0.0.8).

Task 1:

We have relaxed the thresholds (a lot). Now the thresholds correspond to the maximum differences obtained from the ClubFritz (not the 3rd quartile). See the description document.
We are now also only asking that 64 out of 80 HRTFs comply with the thresholds (instead of all 80 HRTFs previously).
Two more (minor) comments:
- The SCUT collection has been acquired at multiple distances. The classifier already selects the farthest distance. Now, the validation code is the same.
- Small bug: We have added the absolute function from the metric difference computation, as we did for the threshold computation.

Task 2:

We have updated the description document to say that the upsampled .sofa files should be saved in the same format as the original files—float64 (not 24 bits, which was wrongly stated perviously).
We also realised that the ITD estimator in the ‘Spatial Audio Metrics’ Python library was not low-pass filtering the inputs before performing the autocorrelation (this is applied by default in the Auditory Modeling Toolbox and should have been applied here). This low-pass filtering operation has now been added to the latest version of the ‘Spatial Audio Metrics’ library (v0.0.8). This additional filtering will affect your ITD results, so please pull the latest version of the ‘Spatial Audio Metrics’ library.
The ITD threshold has also now been increased to 100 μs (from 62.5 μs) due to the new low-pass filtering operation (mentioned above) increasing some of the ITD errors. The plots in the description document have also been updated to reflect this change.
We have also released the code used to create the downsampled HRTFs so that participants can check that their approach is working by downsampling one of the publicly available HRTFs in the SONICOM dataset, which they can then use to validate their upsampling methon (https://github.com/Audio-Experience-Design/LAPChallenge/blob/main/task2/task2_create_sparse_hrtf.py).