ShanonPearce
commented
3 days ago The results after EQ look like a strong improvement, especially in the low frequencies which is good to see.
I am no longer maintaining this dataset, however I am maintaining the ASH Toolset project (https://sourceforge.net/projects/ash-toolset/) which can be used to generate customised BRIRs. I came across the same problem with large variations in the low frequency response when putting together the tool. The low frequency variations have now been significantly corrected in the tool, so it should be less of an issue moving forward for BRIRs generated using the tool.
I've been having this issue for a while but haven't really mentioned it at all, I kinda just made my own solution. The Impulse responses do not have a flat Frequency Response (which is imo pretty important for an IR). This type of convolution is, afaik, usually very sensitive to peaks as well as dips and having a 5db peak at 50hz causes pretty bad auditory masking. I've been kinda fixing these myself ever since I used any type of HRIR (this isn't exclusive to ASH dataset) by using the built-in EQ from HeSuVi since it was a pretty fast workaround and it always sounded much clearer. I feel like if we could potentially add an EQ'd set of these HRIRs based off of a +/-2db flat target with around 3db downward slope tilt it might improve the experience for some that might be dissatisfied with the original ones. The image shows a channel from TV_Studio matched to my previously mentioned target compared to the original frequency response using REW.