Specifically, AL_[EAX]REVERB_ROOM_ROLLOFF_FACTOR always applies to an inverse
distance rolloff model, where the source's AL_ROOM_ROLLOFF_FACTOR applies to
the selected distance model. Consequently, the normal distance and cone
attenuation only needs to be calculated once for all sends, with the effect's
room rolloff calculated and applied separately.
The idea here is that the initial reverb decay can't become less than the dry
path distance attenuation, as the dry attenuation represents the audio that has
not yet had a chance to start reflecting in the environment. As well, the
reference distance indicates where there is no distance attenuation, with any
initial attenuation set by the environment itself.
So what we do is use the dry path attenuation as the baseline for what's mixed
to the reverb, with the decay rate indicating how much of the remaining room
(non-direct) energy attenuates with distance.
This may be over-complicating it. Other sources hint at a more typical XdB per
doubling of distance, with X varying depending on environment properties (room
size, absorbancy, etc). This could be handled by applying a normal inverse
distance attenuation model, with a rolloff factor generated from the reverb
properties (density, decay rate, etc). Will need more testing and research.
Directly mixing a non-UHJ stereo signal into a UHJ stereo signal results in a
non-UHJ stereo signal. Such a mix can't be properly decoded anymore.
An option can probably be added for users that intend to listen to UHJ output
undecoded and let a stereo sound come through as-is on their speakers, but it
probably shouldn't be the default for cases where the output may be decoded
back.
It's not available as an AL buffer format (yet) since I'm not sure how to
expose it. Internally it seems fine as a separate channel configuration, but
because OpenAL combines the channel configuration and sample type, a flag may
work better there.
It's inconsistent behavior with spatialized B-Format sources, and causes a very
apparent gain disparity between such a source on the listener compared to just
barely off-origin. It's also not something an app can easy correct for if it
doesn't want that behavior, since the source gain clamps to 1 by default. In
contrast, an app can attenuate sources as they want with little issue.
It's too unstable with larger vectors. Even when the source and listener
positions are the same, floating point precision can cause noticeable rounding
errors.
Rather than stopping voices/sources when the device becomes disconnected, the
context can be set to leave them alone. As a consequence, their state will
remain as playing and they'll keep their last known sample offset indefinately.
For applications mindful of this behavior, it will allow resetting or reopening
the device to reconnect and automatically resume where it left off.
This is mostly for the SampleConverter, used by some capture backends. When
recording at really low rates, like 5512hz, with a device capturing at a higher
rate like 44100hz or 48000hz, it hits the filter's downscaling limit and
produces pure silence.
In such cases, it's better to just accept some aliasing noise so that the app
will still get some recognizable audio. The alternative would be to scale the
desired rate by 2x, 3x, etc until it's above the bsinc limit, then take every
2nd, 3rd, etc sample of the result as if by an extra simpler resampler pass.
