libera/#shirakumo - IRC Chatlog

19:44:02 Colleen <SAL9000> reverb effects usually introduce latency -- it's still RT-ready if each processing step (N samples in, N samples out) completes within deadline, no?

19:44:25 Colleen <shinmera> Well yeah

19:44:42 Colleen <shinmera> My worry is that, with that much auxiliary data, it won't complete within the timeslice.

19:45:04 Colleen <SAL9000> How big are your timeslices?

19:45:26 Colleen <shinmera> depends on the latency you allow :)

19:46:02 Colleen <SAL9000> well, the "pro audio" guideline is 10ms round trip

19:46:06 Colleen <shinmera> iirc in Kandria it's defaulted to 5ms buffers, but the actual time you have to fill the buffer in is of course much less than that since there's much other work to be done, too (decoding, mixing, playback)

19:47:02 Colleen <shinmera> wait, no, 0.05s buffers in Kandria, so 50ms?

19:47:03 Colleen <shinmera> I can't brain

19:47:24 Colleen <SAL9000> 0.05s is indeed 50ms

19:47:34 Colleen <SAL9000> but that doesn't necessarily mean 50ms of latency

19:47:56 Colleen <shinmera> sure.

19:49:04 Colleen <shinmera> I'm also intending on adjusting the reverb parameters at runtime. Wonder if that's feasible, too.

19:49:17 Colleen <SAL9000> watch out for "zipper" effects? not sure if that's the right term.

19:49:19 Colleen <shinmera> (I know it is in general, talking about this implementation)

19:49:26 Colleen <shinmera> SAL9000: I have no idea what that is

19:49:38 Colleen <SAL9000> I haven't poked into this area for a while, so my memory is spotty

19:50:00 Colleen <SAL9000> a trivial example: say you have a volume control (with unlimited slew rate)

19:50:10 Colleen <SAL9000> your input is a sine wave, for simplicity

19:50:34 Colleen <SAL9000> if you apply volume adjustments at anything other than a zero-crossing point, you risk creating an audible discontinuity if the slew rate is high enough

19:50:47 Colleen <SAL9000> (yes, I've observed this in practice)

19:50:47 Colleen <shinmera> sure.

19:51:26 Colleen <SAL9000> delaying the input adjustments is "enough" for simple stuff like gain or eq -- not sure how to "de-zipper" something like a reverb, though.

19:51:54 Colleen <shinmera> you smooth the parameter change itself.

19:52:03 Colleen <shinmera> using something like https://www.musicdsp.org/en/latest/Filters/257-1-pole-lpf-for-smooth-parameter-changes.html

19:52:19 Colleen <SAL9000> I guess capping slew rate is a reasonable answer.

19:52:47 Colleen <SAL9000> You can treat e.g. wet/dry mix as 2x gain and use step-at-zero-xing for those.

19:53:01 Colleen <SAL9000> the more complex reverb params you'd have to smooth though.

19:53:20 Colleen <shinmera> https://github.com/velipso/sndfilter/blob/master/src/reverb.h#L263-L282

19:53:41 Colleen <shinmera> I'm quite sure this implementation is overkill and I could get away with something much simpler.

19:53:55 Colleen <SAL9000> probably

19:54:05 Colleen <SAL9000> I'm guessing you want "indoor/outdoor" reverb fx for kandria?

19:54:27 Colleen <shinmera> I want reverb that changes depending on how "tight" the environment is based on raycasts.

19:54:39 Colleen <SAL9000> hmmmmm

19:54:47 Colleen <shinmera> As shown here https://www.youtube.com/watch?v=BrUQdd96qzk&t=1307s

19:55:06 Colleen <SAL9000> I wanted to bring up IR-convolution as an option.

19:55:16 Colleen <shinmera> Yeah nah

19:55:29 Colleen <shinmera> Making /that/ smooth is even more difficult :)

19:55:51 Colleen <SAL9000> It's more suitable for a small set of environments (small space, large space, outdoors), but it could work if you generate new IR's based on raycast data?

19:56:24 Colleen <SAL9000> Smoothing between different IR curves could be done as a crossfade between 2 different convolving filters

19:56:26 Colleen <shinmera> Generating IRs is a very heavy process.

19:57:01 Colleen <SAL9000> Set of pre-generated IRs + mixing of IRs and/or wet/dry mixing?

19:57:07 Colleen <shinmera> Listen

19:57:08 Colleen <SAL9000> Just throwing things at the wall here :-)

19:57:16 Colleen <shinmera> I'm sure I can get something done like shown in the talk

19:57:35 Colleen <shinmera> The only thing i'm unsure about aside from the exact parameters is that one triangle thing in the diagram.

19:58:14 Colleen <SAL9000> "Collision Rate/Material"?

19:58:31 Colleen <shinmera> That's illustrating how the parameters are changed

19:58:53 Colleen <SAL9000> I suspect that one's a gain block.

19:59:04 Colleen <SAL9000> i.e. "how much does that surface deaden sound reflections?"

19:59:31 Colleen <shinmera> hmm.

19:59:50 Colleen <SAL9000> Delay -> Gain -> Low-Pass Filter -> All-Pass Filter

20:00:05 Colleen <SAL9000> (tbh I still don't grok what all-pass means and why one wants it)

20:00:27 Colleen <SAL9000> ohhhhhh.

20:00:50 Colleen <SAL9000> "Collision Rate / Material"... as in, the combination of "how many rays hit this surface" and "what kind of surface is it", perhaps?

20:00:55 Colleen <shinmera> yes

20:01:15 Colleen <SAL9000> more rays = louder, stronger deadening = quieter, thus the division

20:02:00 Colleen <shinmera> As for why all-pass, this article explains a bit https://www.earlevel.com/main/1997/01/19/a-bit-about-reverb/

20:03:08 Colleen <selwyn> back 2 the 90s

20:03:09 Colleen <shinmera> Anyway, I /am/ going to port that crazy reverb implementation, and will then prob do something custom for kandria.

20:07:03 Colleen <SAL9000> aha, so APF basically rotates the phase of the signal w/o changing the amplitude

20:07:09 Colleen <shinmera> ye

20:07:13 Colleen <SAL9000> makes sense now :-)

20:07:44 Colleen <shinmera> kinda. I'm still struggling to get my head around the concept of 'phase' for PCM.