Best sound quality - run 96khz samplerate, do not oversample.

[tony tony chopper]

My target medium is always either good quality VBR mp3 (for internet) or cd 44.1 / 16

If your target is 44 and you're mixing at 48, you really have to be sure of the resampling involved. Don't trust the crap that's in the average CD burner, & don't trust all audio editors either.

[Doug1978]

^^ Many thanks Tony for the helpful reply.
I'm happy to let Voxengo R8BrainPro handle all my resampling duties, as I think it's pretty good.

However, do you (or anyone else) think there's that much of a difference between 44.1 and 48, in terms of avoiding any artifacts, to validate choosing 48 at the start please?

I suppose I could just stick to what I'm used to, 48 up to 96 Nebula - down to 44.1 via R8Brain Pro.
But is the 48 part just overkill please compared to 44.1 at the start prior to the final mix?

Many thanks

[tony tony chopper]

However, do you (or anyone else) think there's that much of a difference between 44.1 and 48, in terms of avoiding any artifacts

As you could see in this thread, some are gonna say yes.
Just consider that 48khz is just around a semitone above.

[Doug1978]

Cool - thanks Tony.
I've made my decision: switching to 44.1 khz saves me approx 4% CPU on this laptop with just one instance of Synthix or Zebra playing!
I'll go with that at the risk of a semi-tone more's worth of artifacts creeping in.

[aciddose]

However, do you (or anyone else) think there's that much of a difference between 44.1 and 48, in terms of avoiding any artifacts, to validate choosing 48 at the start please?

assuming the anti-aliasing filter is the same at both sample rates and located at 20khz in both cases, there will be 2x as much cut before nyquist at 48khz as 44khz. 20 ... 22 (2k) or 20 ... 24 (4k).

of course in most software you'll find the filter located at nyquist, not at 20khz. in which case the difference will be that you'll still have 2x as much cut, but the measurement will be only that you'll have a doubly reduced level of aliasing at 20khz, not anything to do with nyquist.

although this difference exists, the end result is the same. you'll have much more cut of aliasing with the slightly higher sample rate at the frequency of interest. (20k)

by the way, try switching to 11.25khz, or if you want to be more practical, use 40khz and save another percentage. (even more than 48k vs. 44k, 40k vs 44k is 10% rather than 9%.) then upsample your final output to 44.1k using an interpolation.

remember that if you have say a 80db/o anti-aliasing filter, it means that your aliasing will be at -80db maximum at 20khz if your sampling rate is 80khz. if your sampling rate is 44.1khz, you'll only have a reduction of 10db, assuming a perfect linear slope. actually it will be worse. (edit, i don't want to both calculating the actual value at that position in a typical slope because it is too damn variable. i'll just change it from "much worse" to "worse". you can assume it'll be pretty close to 10db though assuming that this time i did the math correctly. in any case, 2k is only 10% of an octave from 20k, which would be 40k requiring a sample rate of 80k.)

[Doug1978]

Very informative (and intriguing) post AcidDose.
Thanks for the info - I'll do some further experiments

[Z1202]

Not having time for reading the entire thread, I'd like to throw in a few thoughts, possibly reiterating some of the mentioned ideas.

Using 44kHz vs. 88kHz for sound recording and reproduction is still questioned by a number of people (including myself ), but the differences, if any (no reliable evidence exists so far), are really minimal.

The other story is generation and processing. This includes the antialiasing of the oscillators and/or nonlineartities. It's typically never possible to fully avoid aliasing on those, so it's always up to the plugin's developer to decide on the necessary amount of the antialiasing, and this is always subjective. The developer might judge the quality as sufficient, but maybe I don't. Now, if I raise the sampling rate, the aliasing will decrease further, possibly to an acceptable amount. The nonlinear phase response distortion (close to Nyquist) of the IIR filters is less critical, but I'm not aware of any means to fight that without oversampling, so this even doesn't go anywhere unless you oversample.

From a practical standpoint, most of the synth plugins I use do not sound satisfactory to me at 44kHz, but they do so at 88kHz (that's of course subjective, some will find the difference not critical, some won't be able to hear it at all). It's interesting, that the difference is critical for me even with as simple setups as an osc though a swept filter.

Regards,
{Z}

[Jafo]

Oversampling to 96/88.2 only gives you one more octave of clarity at the top -- helpful

an octave fully inaudible, so how is it helpful? Yes, helpful but only if you plan to make it audible (by pitch shifting or resampling down..), but that's for special effects.

"helpful, but no silver bullet" is what I said. Note that "helpful" does not necessarily equate to "a solution." Furthermore, the idiom "no silver bullet" means that it works in some but not all cases. Sorry for any confusion.

[tony tony chopper]

Furthermore, the idiom "no silver bullet" means that it works in some but not all cases.

but which cases? You wrote an octave of clarity, it can't be "clarity" if you can't hear it.

[Fritze]

the risk of a semi-tone more's worth of artifacts creeping in.

Epic!

[@midnight]

Question,

And maybe I'm gonna sound dumb, but

Why not just low-pass everything starting around 18, 19, 20khz? I mean, if it gets rid of aliasing. Surely the grand majority of people can't hear anything that high anyways? IMO topping out at 18khz, its gonna sound exactly the same played back on most speakers?

[bmrzycki]

Why not just low-pass everything starting around 18, 19, 20khz? I mean, if it gets rid of aliasing.

The problem is that it doesn't get rid of aliasing. The effects are spread back across the frequency domain, which is why Urs and others speak of bandpass filters to eliminate the aliases.

Here's an interesting graphical look at aliasing done by some of the guys at discoDSP. It might help you "see" the problem a bit better: http://www.discodsp.com/highlife/aliasing/

[Aleksey Vaneev]

When going from 44.1k to 88.2k sample rate you not only get "1 octave of clarity", you get 2 octaves! Because aliased components travel back on the spectrum after reaching 88.2kHz boundary and they have to travel one more octave to reach the audible band. If you work at 192kHz instead of 44.1kHz you get almost 8 additional octaves for harmonics in comparison to 44.1kHz.

[aciddose]

ah ha! i knew my calculation was %#*@ed in some way. that's how.

so it's yes even more of an improvement just to run at 48k compared to 44.1k, and then again at 96k.

i didn't bother to mention it at all in the case of a filter at 20k, you have to count the reflection as well.

now that i think about it though what i did was was perfectly accurate, just left that factor unconsidered.

[metamorphosis]

I oversample some effects and I run 96khz. For certain things it can make a big difference - as to whether the difference is always desirable, no, it's not always. But the basic premise of this thread is correct. Oversampling is a bit of a hack if you're running at 44khz, and a computationally expensive one compared to just running 96khz from word go. Anyway - this's already been discussed to death, and I trust my ears more than others opinions, so I'll leave it.