I've always used 48k because it gives you 24k nyquist vs. 22.05k. This is close to double, so filters can get that extra space to work before even hitting nyquist, assuming cutoff = 20k.EnGee wrote: Final question: if using 88.2 or higher. Is there a need then to use high quality option (or oversampling) in the synth? As it is already with high sample rate?
I would be interested to know more about this topic but mostly for my projects I will keep using 44.1
Honestly though it doesn't really matter that much. If you have to choose and you have no reason to use 44.1, I'd advise you to use 48 if you otherwise hear no difference.
( 44100hz is stupid: https://en.wikipedia.org/wiki/44,100_Hz )
This is also important though, filter (one pole, 6db) gain at a frequency is equal to:
- low-pass: gain = fc / sqrt(fc^2 + hz^2)
- high-pass: gain = hz / sqrt(fc^2 + hz^2)
So if you move the cutoff to nyquist, the gain at 20k will be higher than otherwise if you use a higher sample rate.
Most hardware uses a low-pass cutoff around 30k to 40k to cut out-of-band (inaudible) content.
As far as whether you need oversampling in the plugin, the plugin's oversampling will be in addition to the higher rate you run and and yes, it may have an influence. You need to decide whether it makes a different on a case-by-case basis yourself. Use your ears! Play higher notes, mixtures of content (breaks with cymbals and snares, vocals, synths) and see if you hear any difference.
The proper way to test a non-obvious difference is called A/B testing. If you repeat the text while randomizing which one you listen to and try to categorize "this is A" and "this is B", you can check how many times you were right.
If you're right more often than 1/2 the time there is a chance you can hear a difference.
If you're right 1/2 the time there is near zero chance since this would be random.
The only way to improve the accuracy is to test many times. Try 4 usually, 8 or 16 when you're really wanting certainty. This proves "is there an effect I can hear?" with reduced error the more times you test. Exactly like flipping a coin, you can always get heads 15 times in a row, so when you want to confirm a result repeat the test, then repeat it again, then again, again and once more. More positive results = higher certainty. If you test 16 times and pick the right A/B every time you're pretty sure the difference is obvious.
Flipping a balanced, unbiased coin for heads 500 times in a row is so unlikely we can call it "near impossible". Not actually impossible, but in any practical sense it is never going to be experienced by anyone with such low odds.
Now about intermodulation distortion, most analog equipment uses low-pass filters at 30k to 40k as I mentioned. These often aren't implemented in software. (They add an expense that is usually not going to make any difference.)
When you over-sample though, you must filter away content above the band you're interested in (0 to ~20k) to reduce this sort of distortion.
Keep in mind though that it requires non-linearity to occur so there is no need to filter between two linear EQ plugins.
Applying the low-pass afterward is exactly the same as before, or between the two. With "linear time-invariant" filters (LTI, see Wikipedia as usual if you want to melt your brain with equations) the order you do them doesn't matter.
This is called the "commutative property" of LTI systems.
