Jonbouy
drewfx1
A few points:
In terms of things like digital amp sims (and also some synths), anytime you create distortion (or certain synth waveforms) you are creating high frequencies which will cause aliasing in digital. But for this reason in the modern world (where CPU power is abundant) any competent programmer upsamples their amp sim/synth plugins internally to avoid this problem. So it's not clear why running at higher rates is of any benefit for modern plugins (assuming whatever "quality" settings they have are set to their highest settings). If it was beneficial to upsample even more, the programmers just should have made an even higher quality setting available. Older plugins may or may not be a different story.
Hi Drew some knowledgeable, relevant good sense as usual. Regarding the above point though did you read the article by Bootsy I linked which covers this particular subject from his view point?
Just read it.
There are basically 2 pieces of interest:
1. He argues that if a plugin upsamples, it has to do SRC's and starts talking about the artifacts/ringing/etc. that this introduces. While technically true, note that not only is there is no discussion of how audible this is in the real world, but he also talks about how good (more CPU intensive) offline SRC is in another paragraph. How bad does the SRC have to be for it to be audible? Or how many times do you have to repeat a better SRC before it becomes audible? What does it sound like? Is this stuff audible only for specific types of high level, high frequency test signals, or under more general circumstances?
IMHO, people who talk about artifacts without bothering to put the audibility in context are not helping anyone. In some cases it may encourage people to believe they hear stuff that isn't really audible (if it's even present). And prevents anyone from being able to make any sort of a reasoned judgment. I only care about artifacts I can hear, so don't just tell that they're there - tell me under what contexts they might be audible and what they sound like.
I do agree that computation wise, it makes sense to avoid unnecessary SRC's and the CPU costs involved, but keep in mind that if you run at a higher sample rate
all of your processing is done at the higher rate, including all the linear processing that gets no benefits at a higher rate. Obviously it will also affect your disk usage and throughput as well.
Also note that for processing that produces lots of higher harmonics, just going up to 2x (i.e. 88.2 or 96k) will be insufficient, so the plugin would still have to oversample and you won't necessarily gain much CPU cost savings (you will move the plug's SRC artifacts higher though - which is a benefit if they were audible in the first place).
Whether doing repeated SRC's produces audible problems is an interesting question. It might or might not, depending on a number of factors. So in theory it might make sense to avoid this. But without doing carefully controlled double blind listening tests involving the specific SRC's in question, it's hard to say.
IOW, there may not be a single obvious best answer for all situations.
But personally I have always argued that a DAW should allow you to set the sample rate at which individual or bins of plugins operate, as that way it could easily eliminate any redundant SRC's.
2. In terms of "filter coefficients", filter warping, IIR filters, etc., that was essentially the point I made about EQ curves changing. Minus all the melodramatics, of course.
In terms of filter stability and the difficulty in achieving certain curves, it's again really just a question for the programmer. If you can't get what you want in a filter at a lower sampling rate, you just upsample as necessary.
So there's really a couple of distinct overall points here:
There are places it's clearly beneficial to do processing at a higher rate. Programmers can/should/will do this as necessary. Generally they sort of know what they are doing and do it. But older plugins (that were trying to save then more limited CPU) might not use a higher rate even if ideally they should.
and
What is the impact of repetitively doing SRC's at each plugin vs. running at a higher sampling rate throughout? How good is the SRC in each plug and how many times does my audio go through them? What are all the costs and benefits of doing this in different ways? Does one approach make sense universally, or does it depend on what you're doing? Are we making these decisions based on theoretical worst case scenarios (perhaps using carefully chosen high frequency test signals to bring out artifacts), or based on more typical real world audio?