The Easy Communications (EasyCom) dataset is a world-first dataset designed to help mitigate the *cocktail party effect* from an augmented-reality (AR) -motivated multi-sensor egocentric world view.
Dear EasyCom team,
thank you very much for your nice dataset.
I am working on you ATFs measurements. I just realized that there are two h5 files: Device_ATF.h5 and Device_ATF_v5.h5.
In the former, the coordinates looks distributed on a regular uniform 2D mesh, while in the latter, the distribution along the elevation axis is non-uniform.
Could you provide some more details about this? Did you simply collected new measuments?
I am trying to download the dataset via the command while I keep getting the error Connection closed at byte 802228. Retrying. I cannot download manually from the release page either. I have tried two different network but issue remains.
Hi,
We are working on our graduation project, as part of the project we are trying to understand several things in your article. We would be happy if you could answer some questions:
1.What kind of classification do you do, and how did you deal with the large number of classes? Is there use of multi-class-multi-label?
2. How do you deal with the fact that the coordinates are relative to a fixed place in space? (In the article it was written that the coordinates are relative to the left leg of the table). If we assume there is a speaker in a fixed place, we will get a situation where his labeling is fixed, but the array is constantly moving, so the position relative to the array changes (while the labeling remains constant).
Would it be possible to share a PyTorch Dataset class that handles the preprocessing of the data and the data loading? That would make working with the dataset very straightforward !