Sync AudioServer & AudioDriver buffer sizes and fix latency inaccuracies #52626
Conversation
ellenhp
force-pushed
the
latency_fixes_4_0_dev
branch
4 times, most recently
from
d50d934
to
f545022
Compare
Co-authored-by: Ben Armstrong <ben@avsoftlabs.com>
ellenhp
force-pushed
the
latency_fixes_4_0_dev
branch
from
f545022
to
e82d9fe
Compare
akien-mga
added
the
cherrypick:3.x
|
Reduz mentioned in rocket.chat that we should be talking to the audio system in each driver to determine the latency, and that this method of calculation isn't accurate enough to be acceptable. I'm considering splitting this into two PRs. One to make the buffer size impacted by the latency parameter, and the other to make get_latency behave correctly. The second goal will be much more difficult for me, since I can't really test all platforms easily. |
| @@ -1345,7 +1345,14 @@ void AudioServer::init() { | |||
| channel_disable_threshold_db = GLOBAL_DEF_RST("audio/buses/channel_disable_threshold_db", -60.0); | |||
| channel_disable_frames = float(GLOBAL_DEF_RST("audio/buses/channel_disable_time", 2.0)) * get_mix_rate(); | |||
| ProjectSettings::get_singleton()->set_custom_property_info("audio/buses/channel_disable_time", PropertyInfo(Variant::FLOAT, "audio/buses/channel_disable_time", PROPERTY_HINT_RANGE, "0,5,0.01,or_greater")); | |||
| buffer_size = 512; //hardcoded for now | |||
| if (AudioDriver::get_singleton()) { | |||
There was a problem hiding this comment.
I think it should be always assumed you have an audio driver at this point, so the check is unnecessary.
|
I think the general idea for this PR is good, and it could be merged as is, but I have the feeling that
I have the feeling that this needs a bit more of research to ensure its really done as best as it can possibly be. |
AThousandShips
removed
the
cherrypick:3.x
This is a rebase of #38280 with some style fixes and slight changes all around. More accurately I reapplied all changes by hand, so if you've previously reviewed that PR please still look carefully at this one.
Please review paying special attention to any assumptions that may have been made about the buffer size in the AudioServer being 1024, and more recently 512. According to Ben and another developer in the comments of # #38280 it sounds like this change or some variation of it has been used in at least one shipped title, but I worry that (as an example) perhaps there's an FFT in one of the AudioEffects that will break when passed a non-power-of-two by an AudioDriver that chooses to only allocate 500 samples instead of 512, or something.
Overall this change will be an incredible help for developers that need low-latency audio, since before the AudioServer limited latency to at least the length of its hardcoded buffer.