[viccis]

If you try to use empiricism when it comes to certain groups audiophiles, you are going to be sorely reminded that it's basically the equivalent of healing crystals for a different type of person. 24/192 is useful for mixing/mastering, but completely unnecessary for the end product to distribute for listening.

[evo]

24/192 is also great for digital synthesizers--if you're generating a waveform like a sawtooth that has theoretically instantaneous transitions, they can eat as much frequency as you can give them. Running at 44khz loses noticeable high-end content.

Most modern digital synths have already caught onto this and run internally at much higher sampling rates even if their output gets downsampled, but sometimes you run across a vintage plugin that runs at the host audio rate and working in a higher sampling rate is audible.

[Applejinx]

Hydrasynth aliases like a mad thing. My flagship synth ended up being Summit, and its oscillators are digital but run at a crazy high sample rate. Did likewise with some Chord Organ modules: that Teensy board it was built on could do chord audio at 300k and over a megahertz if you were just generating one wave as simply as possible. The freedom from aliasing really helped the sound, for all that it's a 12 bit analog output. A squarewave is a 1 bit signal…

[nullc]

> 24/192 is also great for digital synthesizers--if you're generating a waveform like a sawtooth that has theoretically instantaneous transitions, they can eat as much frequency as you can give them.

So if your synthesizers do not use proper band-limited oscillators then 192KHz is _FAR_ too slow. You'd want to be running at hundreds of KHz, perhaps a few MHz.

In reality synth software that doesn't sound like crap uses band limited oscillators and should work okay at 48KHz too. That said, even if the oscillators are band limited it may be the case the varrious modulations aren't band limited properly, as getting those wrong won't sound instantly wrong (in particular because you have to modulate to make it wrong, and the underlying change of the modulation may make it harder to tell its wrong).

Though also in those cases if you're not counting on every step being properly band limited then 192KHz may be an improvement but you're still probably getting some meaningful aliasing. I think given how fast computers have become relative to digital audio there is probably a good case to just make any "modular synth" run at 32-bit 480KHz or even 4.8MHz through every stage that could process the audio.

Maybe 192KHz really is enough to suppress the aliasing artifacts but I think to be convinced of that I'd want to see a system that supported both and validate that the difference between a downsampled 48KHz output from the two modes was below -90dB or something.

Or otherwise you can just declare that the aliasing is part of the sound and then there are no right choices... 24khz sampling, 48k, 192k ... who cares, use what you like best. :)

[Archit3ch]

> I think given how fast computers have become relative to digital audio there is probably a good case to just make any "modular synth" run at 32-bit 480KHz or even 4.8MHz through every stage that could process the audio.

1. It should run at FP64 if you want to preserve filter resonances, etc.

2. At 10x/100x fixed-rate oversampling, even a modern "fast" CPU will have very few cycles per (higher-rate) sample to run the DSP for 1 "module" of the software modular. Forget about interconnected modules, multiple tracks, or polyphony. For this kind of "analog"-style processing, it's better to run adaptive-rate algorithms (think SPICE) instead of wasting compute on unnecessary extra audio samples.

[nullc]

so an 8-core zen4 should be able to sustain more than 300 gflops of 64-bit multiply-adds. At 480khz that's 625k operations per sample. I'll grant the 100x oversampling was probably too ambitious. :P

For adaptive rate I think the issue there is you have a hard-realtime constraint for this usage (even if you wouldn't mind rendering offline, you kinda have to hear it realtime to tweak it-- after all you might tweak it in a way that brings out an artifact you like and then be disappointed by the render). Also in the case of a whole modular system having all sorts of different parts needing to be part of the adaptation loop seems pretty hard to me.

My thinking was just in general that 192k is really not enough to prevent aliasy algorithms from messing up. If you are alias safe you can probably run at 48k and be fine. If you're not, you really want to go much higher.

[Archit3ch]

I'm travelling without my Zen 4 machine, or I could test it. ;) Oh well, Compiler Explorer is enough to look at these microbenchmarks on your own.

These simulations are single core to avoid core-to-core latency. Number of cores isn't relevant unless you want to run independent voices/channels and sum them at the end.

So you start with a very optimistic ~90 GFLOPs of 64-bit FMA on Zen 4. Unfortunately, not all operations are clean multiply-adds. Realistically, you'll need trigonometric functions and LUTs, which are quite slower. Btw, the tradeoff between when to compute vs LUT is very fragile and can change due to a ton of factors (notably integrator algorithm).

Then the data you are operating on won't fit cleanly in AVX-512 registers, requiring spills to L1 cache. Ok, still fast on a modern core.

Of course, the peak theoretical number assumes clean vectorization with double-pumped AVX-512... which also won't happen in practice. Classical DSP will fare better (https://www.youtube.com/watch?v=Ssq0a-YdamM) but SPICE integrators are inherently branchy and divergent. Especially for adaptive integrators, you'll waste a lot of operations trying to "lock in" at the exact time point where the waveform turns a corner. Apple Silicon is better at this messy, branchy code.

So yeah, it's possible-but-hard to hit hard-realtime under these conditions.

[Blackthorn]

You can generate perfect band-limited sawtooth waves at 44.1khz, there are multiple techniques for doing this and most production digital synthesizers use them.

Oversampling gives you headroom for aliases for the rest of the synth that is more vulnerable to it.

[evo]

Yeah, I was oversimplifying a blit, the raw waveforms are usually okay, but I distinctly remember old-school VSTs where you couldn't achieve a nice saw lead at 44.1.

[Blackthorn]

It's tough to tell without specific names, but I imagine a lot of particularly old* VSTs were written to use naive sawtooths rather than perfect band-limited ones, which would have terrible aliasing at 44.1 khz. Oversampling those would help a lot!

* Some people are still making this mistake, despite information on the (many) ways to do it the right way being widely and freely available!

[evo]

I wonder if there's also distortion or ring modulation stages where some of the energy above hearing range might spill into audible sidebands if they're not nyquist-limited first.

[Blackthorn]

Yeah, that's the "rest of the synth" part that's more vulnerable to aliasing.

There's some ways to do band-limited distortion but...they aren't nearly as widespread, easy, or universal as band-limited oscillators.

Ring modulation is funny though because you'd ideally want the sidebands to modulate down by default rather than filter them out, that's why you're using it.

[dist-epoch]

No synth generates sawtooths by literally drawing a saw tooth in PCM. The distorsion you get if you do that is not subtle at all.

[colmmacc]

32-bits are great for recording too because they do an incredible job of capturing the dynamic range without having to be precise on the preamp settings. It removes an entire job from the recording workflow.

192 for mixing and mastering can be useful especially if you're doing a lot of effects, especially anything that pitch shifts. But I've seen low quality phone-microphone recordings make it to the master; if you capture lightning in a bottle, it hardly matters what the settings were, what the microphone was, or anything else.

[PaulDavisThe1st]

The limit on current DACs is 18-22 bits. The rest is just brownian noise. Literally.

[Aldipower]

Even with mixing/mastering 96khz is enough for persisting to files. But as another commenter said, 192 is useful, if you bend and stretch samples!

[tshaddox]

They literally sell actual crystals that you’re supposed to place on top of speakers and amplifiers to make them sound better.

[Blackthorn]

We had a really nice crystal decoration that I happened to put on top of one of my TV speakers and, wouldn't you know it, it had this resonant frequency somewhere around specific human speech frequencies that drove us absolutely bonkers until I figured out the cause and moved it.