[1604] HRTF deadzone at low angles
tracked
The Heavynator
audio sources below the listener exibit a large sudden transition between filter states when passing directly over or just to the sides of them (even with bilinear filtering) due to what im assuming is a "hole" in the HRTF data.
providing a custom .SOFA file can "fix" this pointing to a problem with the defult HRTF.
an example can be found here https://vrchat.com/home/world/wrld_cd04125f-3c36-43f9-93c4-999674297552. (turn right from spawn and cross the bridge while looking straight ahead)
I should mention however this may not be fixable on VRCs end since this appears to be an inherent issue with steam audios HRTF as this occurs in edditor and in valves own games like CS2 but i thought i'd at least bring attention to this issue regardless.
Log In
a
alareis
Doesn't seem to be related to extreme angles. It persists at any point, any distance.
Is there some specific setup you use with those audio sources? I've been trying to replicate this with existing sources and clips, but so far no luck.
Seemingly the only ones affected are the sources for waterfalls and waterway "steps" in that map.
Ambix clips behave odd at worst, but nothing to the demonstrated degree.
The Heavynator
there's nothing special about that source other than it not using a vrc spatial source component (since i find that a little cumbersome vs the spatialize toggle).
It could be that the sound for the waterfall is simply the worst case senario for the spatialiser since it happens to the AC units in the map too but it is significantly harder to notice (again even with bilinear filtering enabled).
this clip has all other sounds disabled.
a
alareis
The Heavynator: so, after some testing the following seems to be the conclusion:
- Wide-band mono tracks treated as binaural (stereo HRTF input) are mixed down at runtime have azimuth banding caused by destructive interference, judging by the source code; although mono tracks should not be used as such. The wider the band, the more audible are the steps.
- Properly HRTF-filtered clips still have a ~200° phase step; this looks like an internal issue, as no third party sofa profile got rid of it in my testing. I suspect an issue with dual lobe filter, especially given how non-HRTF sources have a panning hotspot, which seems to be related to the L2 ring in listener's relative position calculation. Thus far the issue can only be masked with various effects SA provides.
A difference demo can be seen here, with your name on HRTF-filtered and non-filtered zones (to the side of the "room"): wrld_aa458f70-972b-494b-9eaf-6f6b1483ca9b
Do you have a spectrogram of the waterfall clip by any chance?
To get rid of the azimuth banding on mono tracks I'd suggest turning them into stereo tracks with up to ½ phase offset, seemed to have worked for a couple of tracks I tested with.
The lobe step mitigation can only be tested in the editor at the moment, but it's worth a shot: add traced reverb and prebake reflections. Banding will still be present, but it should be a lot less audible with post-effects being present.
StormRel
tracked
Salbug
Videos Heavy shared showcasing the issue.