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

[tony tony chopper]

It's helpful for processing headroom.

It may be helpful for some plugins, and is better per-plugin. But don't you think that the plugins that need it, are already doing it?

I mean if you open your fav drawing app, and wanna draw a circle, it's gonna draw a properly antialiased one, it's not gonna require you to work at a very high res.
But yes, in some cases it's not gonna do a good job at it (even Photoshop & its shitty 4x-oversampled unhinted font rasterizer), & it can be useful to work at a higher res (but not for everything. And that wouldn't even improve Photoshop's poor font rasterizer since it will still lack hinting. Actually, in a proper editor that would offer font hinting, working at a higher resolution to rescale later would be WORSE. But ok this is for images)

[khanyz]

It may be helpful for some plugins, and is better per-plugin. But don't you think that the plugins that need it, are already doing it?

That was my discussion with Urs, but from my work (which isn't audio) that wasn't the case. It was better to keep a constant samplerate and filter between processes. It minimized calculation errors/artifacts.

Now that I'm doing some audio DSP, I'll be checking out the maths, mainly to check how much of my work at higher frequencies is applicable. At the moment, I would always defer to Urs's expertise.

Either way, there's no may about some plugins working better at sample rates which is much higher than would be justified for just the range of human hearing.

And about it being per-plugin, that would depend on the oversampling implementation and I think that a consistent one would be better. If that means let the host handle it (i.e. keep a constant higher rate) then where's the harm.

As for your drawing metaphor, when you draw a circle the app doesn't store every point on it (raster), it stores meta-data about its attributes (vector). When you view it, that is just quick rasterization (dither) to give you a preview. You never work in raster if you don't have to, you convert to it at the last moment.

So, what are you doing? You are working at the highest quality whilst processing and then reducing it to the destination format at the last point.

Now how would you do that in audio?

Cheers,
Nigel

[aciddose]

It may be helpful for some plugins, and is better per-plugin. But don't you think that the plugins that need it, are already doing it?

the problem is "oversampling" isn't something with a strict definition. there are countless numbers of filters which are optimal for various purposes. going along with your example of image processing, i was just rendering some 3d models today. various filters will emphasize sharpness while introducing a little more aliasing. using an "ideal" filter in terms of aliasing the results came out very blurry, as expected. when using a simple box filter however the results were much better.



although some aliasing is noticeable, over-all the improvement in sharpness without introducing ripple is favorable.

typically it is cheaper to use a filter with less desirable properties. when you run completely at a higher rate and then decimate the result, you won't need to run the filter hundreds of times, multiple times per plugin. instead you'll only need to run it once for every final output (master) channel and you can then select a filter which is far better suited to your needs.

there isn't any harm in using higher rates. i think everyone should be using at minimum 48khz these days. 96khz is preferable, with 192khz being "overkill" in most situations.

when you store data, use the minimum rate of 48khz, 24bit - except in cases where higher resolution is needed during further processing. if storage is an issue, 16bit can be acceptable, although storage doesn't tend to be an issue and generally full rate and floating point are used instead. (which is essentially equal to 24bit.)

if people are using my plugins at 44.1k, they are going to get poorer results. the difference won't be huge, but it will be there. if they use them at 96k, they'll get better results than i do since i run at 48k.

running my plugins at 96k (at least, the ones without oversampling) will be less costly than turning on oversampling. you must take into account that running at a rate which forces oversampling will include many upsample and decimate stages in series through the chain. if the chain instead is run at the higher rate with oversampling disabled and decimated at the end, this will have an equal cost for processing, but lack the needless in-between low-rate stages.

one effect that does need to be kept in mind though is that you will then not be filtering out content above the "master" nyquist, which means more high frequency content could be introduced through-out the chain.

[james0tucson]

[quote="A.M. Gold"]Does any of this matter, though, when we all deliver any audio we produce as mp3's (or worse) or, at best, "CD quality" audio format?[/quote]

Plenty of people deliver FLAC. I'm sure there are still people "making CD's" these days, but why should anyone drop his own standards because of people who insist on remaining in the "media" era? Makes no sense, unless it's a money thing.

[aciddose]

let's take as an example the fact that blu-ray's default and most often used format is 48khz 24bit 5.1. you won't likely see demand for any higher rate. young people can sometimes hear signals above 20khz and this even occurs once in a while in more mature individuals. it's extremely unlikely that you'll need a rate supporting more than 24k, though, since this will cover the extra 0.9999% percent of the market where 40khz would already have covered 99%.

there are reasons for a higher playback rate though which include jitter, anti-imaging filters and other issues although those aren't as important today as they were in the past, 48k is still typically justified in place of 40k.

it makes sense to aim for the highest common denominator of the market, not the lowest. plus, the decision is out of your hands - the biggest, most successful players in the industry have spoken, and their word is 48k.

edit: actually, now it seems "trueHD" lossless format is becoming far more common which uses 96k 24bit 7.1. in my opinion this is crazy, but whatever. now you can see all sorts of neat ultra-sonic properties in these recordings. your cats will finally be able to appreciate your movie collection as much as you

[tony tony chopper]

As for your drawing metaphor, when you draw a circle the app doesn't store every point on it (raster), it stores meta-data about its attributes (vector). When you view it, that is just quick rasterization (dither) to give you a preview. You never work in raster if you don't have to, you convert to it at the last moment.

let's not debate about vectors vs bitmap, both have their pro's & con's. Which is why Photoshop isn't Illustrator.

[A.M. Gold]

edit: actually, now it seems "trueHD" lossless format is becoming far more common which uses 96k 24bit 7.1. in my opinion this is crazy, but whatever. now you can see all sorts of neat ultra-sonic properties in these recordings. your cats will finally be able to appreciate your movie collection as much as you

Yea, here again we have the "bat hearing" analogy invoked.

But this isn't a question of hearing a sine wave played at 25 KHz, it's about whether people subjectively think the same track rendered at 96 sounds somehow "better" than one rendered at 44.1 or 48, in a blind test.

Since it's past 1 am here, I'm not going on a Google hunt, but it's going to come down to the results of such tests, not theories on what measurable (at this point) audio components are relevant to such a question.

If people can hear a difference (and no, I don't mean the freakish .001% who can clearly hear the sine above 20+K) then that is the end of the argument. If so, then there must be something going on that standard analysis and standard models of sound and/or hearing must not be accounting for.

[tony tony chopper]

oing along with your example of image processing, i was just rendering some 3d models today. various filters will emphasize sharpness while introducing a little more aliasing. using an "ideal" filter in terms of aliasing the results came out very blurry, as expected. when using a simple box filter however the results were much better.

Sure you can have good or bad resampling, but it's still not a good reason to process 10 plugins at a high samplerate just because 1 of them would need it. Unless it really ends up sparing CPU, then why not?

there isn't any harm in using higher rates

maybe, but only on those computers from 2050 that can process anything you can dream of

when you store data, use the minimum rate of 48khz,

except that if your stuff ends up being used at 44khz, you will really have to pay attention to the resampling method (& waste CPU on it)

running my plugins at 96k (at least, the ones without oversampling) will be less costly than turning on oversampling.

less costy only for your plugin, more costy for the rest in the chain

if the chain instead is run at the higher rate with oversampling disabled and decimated at the end, this will have an equal cost for processing, but lack the needless in-between low-rate stages.

& less quality

Imagine this: your plugin is processing at 48khz, with 2x oversampling. The output of your plugin will be 48khz, so it won't have content over 22khz. Then later in the chain there's an effect that too has its own oversampling, it oversamples, now it's processing a *96khz input that has nothing above 22khz*.

Now take your scenario, the whole chain is 96khz. You output something that has content up to 48khz, right? So the effect in the chain is now processing a *96khz input that has content up to 48khz*.

Which one do you think is gonna let bleed the most aliasing, for the same oversampling factor?

Edit: I saw you told about that in the rest of your msg. So if you're aware of it, you can't argue that it's better.

[aciddose]

If people can hear a difference (and no, I don't mean the freakish .001% who can clearly hear the sine above 20+K) then that is the end of the argument. If so, then there must be something going on that standard analysis and standard models of sound and/or hearing must not be accounting for.

or that the experiments are not accounting for. such as non-linearity in the audio chain between the signal source and brain (well, including in the brain!). as you mentioned before, phase distortion or artifacts due to various types of filters and so on.

it's very well understood that you can in fact hear tones above 20khz if the product of their modulation is below that point. this effect is used in plenty of real-world products even now.

the problem with the statement that "humans can't hear content above 20khz" is that you must take into account that you're speaking in terms of very narrow undistorted peaks. humans can hear ultra-sonic content just fine if there are products produced that fit into the range we actually measure.

so the issue is not whether it's possible to hear content above some arbitrary frequency range. the real issue is: what changes are occurring to the content below a frequency range that are being heard. how can we measure these changes and quantify them?

one issue is differential bit error in DACs. if the sampling frequency is very high, this error will be spread into a higher frequency range, an inaudible range. there are countless others.

[ariston]

But this isn't a question of hearing a sine wave played at 25 KHz, it's about whether people subjectively think the same track rendered at 96 sounds somehow "better" than one rendered at 44.1 or 48, in a blind test.

Well, we know from a number of studies that low frequencies are "felt" by us (as opposed to heard), and that they have measurable effects on our nervous systems. The same could be true for higher frequencies (I'm not familiar with research in this area). The questions are: are these frequencies a part of our listening experience, and if they are, do they really matter?

I think that, ultimately, any type of recording will always lack the "liveliness" (for want of a better word) of music being played right in front of you. Caring about fidelity is great up to a point, but it, like everything else, can be overdone. A great song is a great song, even on a cheap 10$ radio.

[khanyz]

maybe, but only on those computers from 2050 that can process anything you can dream of

Or the one I have which will happily process enough tracks (~16), busses (~5) and master, all with effects and some with instruments, at 88.2kHz for my usage.

Mind you I did get it in 2005 so you were nearly right.

[aciddose]

Edit: I saw you told about that in the rest of your msg. So if you're aware of it, you can't argue that it's better.

of course i'm aware of that. it's only better if you are aware of it. in my opinion it's much better because the results are more pure.

for example you're going on about eliminating the content below 20khz as if it's some godsend. in reality you're just going to introduce phase distortion and ringing which can throw off a lot of things such as envelope followers. if the original, undistorted signal can be processed or at least minimally filtered, those effects will be reduced even in plugins which do not introduce any aliasing or harmonics by themselves!

for example if you run at 96khz, an envelope follower can apply a 6db/o lowpass filter at 20khz to "window" the frequency range of it's side-chain signal. this will eliminate ringing/peaking outside the desired range without influencing the processing of the audible content. this is much more correct! the ringing wouldn't have been present if we were dealing with an analog circuit for example, as it would have been very rare to see anti-aliasing filters with 90db/o slopes and passband/stopband ripple on par with what we use in plugins today. it would also simply be "more correct" because you'd be dealing with the real "natural" or "naive" peak values of the audible content that you want to be dealing with.

yes there are all sorts of caveats, but this is why i only suggest that 44.1khz is obsolete. my reasoning there is that there is probably no way you'll be producing a finished product at that rate since the native rate of all your modern systems (ipads, blu-rays, etc) is 48khz, not 44.1hz. there is no suggestion that everyone should use a particular rate "just because" with no admission of any caveats. it's just well, a fact that most productions are going to need to be in 48khz minimum from now on and well, that you might as well just get used to it.

i say "96khz is preferred" while assuming that people are not morons and that they're aware there is an expense associated here. although i also posted the fact that 96khz is now the standard rate for blu-ray productions and so you don't have any choice anymore.

otherwise we can all just start using 1hz sample rates and save an awful lot of cpu power on everything.

(or, we can just naively play our sub-rate data into the target rate and allow aliasing and/or imaging to occur since nobody can hear it anyway, right?)

[tony tony chopper]

for example you're going on about eliminating the content below 20khz as if it's some godsend. in reality you're just going to introduce phase distortion and ringing which can throw off a lot of things such as envelope followers.

phase distortion: yes, but you can still go with linear-phase FIRs.

ringing: I believe you have enough between 18 & 22khz for the smoothing (I won't dare saying 16khz because some here probably like to think they hear above that)

envelope followers: the question is do you really want something that you don't hear to pollute your env detection? IMHO envelope following would be a good reason to pre-filter everything.

if the original, undistorted signal can be processed or at least minimally filtered

what's the original signal anyway? Can't a synth process up to 1 billion Hz? If anything above 22khz is so important, you shouldn't discard any important information around 1 billion Hz

yes there are all sorts of caveats, but this is why i only suggest that 44.1khz is obsolete. my reasoning there is that there is probably no way you'll be producing a finished product at that rate since the native rate of all your modern systems (ipads, blu-rays, etc) is 48khz, not 44.1hz.

whether you like it or not, I'd say that the most standard medium for a finished product is a 44khz mp3..

I always work at 44, but 48 is the same to me, the difference is neglectable. As long as people don't keep mixing 44 & 48 without care..

although i also posted the fact that 96khz is now the standard rate for blu-ray productions and so you don't have any choice anymore.

I don't see how a blu-ray target would force you to include anything above 22khz. It's not like anyone would look at a blu-ray's track in a spectrogram & whine because it's actually a 44khz track.

(or, we can just naively play our sub-rate data into the target rate and allow aliasing and/or imaging to occur since nobody can hear it anyway, right?)

where is that aliasing from?

[Doug1978]

Hi all

Sorry for what might appear lazy or a wind-up, but it's honestly not.
My head is just spinning from the previous 6 pages of back and forth viewpoints!


Short version:
Is there any real benefit whatsoever for me mixing at 48khz rather than 44.1khz please?
I have currently chosen 48khz because, as I understand it, it pushes potential artifacts slightly further out of the hearing range of most humans.

I'm on a laptop, so as much as I'd love to do everything always at 88 or 96khz, the CPU hit is just too much for my PC.

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



Long version:

Currently I compose and process at 48khz / 24 bits in Sonar 8 PE.

I sometimes use recordings / samples (recorded mostly at 44.1 / 24) within my 48 / 24 projects, but usually I use vst synths and plug-ins at 48 /24.


When final mixing, I use Nebula 3 Pro at 96khz because of the intentionally different (96khz / 44.1 khz) programs made by Alex B and CDSoundmaster.


I export from Sonar at 48 / 24.
I use Voxengo R8BrainPro to move from 48khz up to 96khz.
I import in Sonar at 96khz and use the Nebula 96khz programs.
I export from Sonar at 96 / 24.

I then use R8BrainPro to move down from 96 / 24 to 44.1 / 16.

I then either keep to this CD format, or I go into Sonar for one final time and use the LAME mp3 encoder to change the 44.1 / 16 into a VBR mp3 file.

Phew!


So should I really be that bothered about the 48 / 24 stage, or should I just stick to 44.1 / 24 (before the Nebula and R8BrainPro stages etc)?
Am I really avoiding any artifacts by choosing 48 rather than 44.1 at the arranging / early vst plug-in mixing stage?
I swear that sometimes I can hear a slight airyness to synths at 48 rather than at 44.1, but this could be my brain playing tricks.


Many thanks guys.

[bezusheist]

if anyone is interested...
i posted a (cough) "test" HERE (kind of a joke) and apparently no one wants to take the bait...
feel free to download the 96k file and play around with it and tell me whether or not you hear a difference when filtering out frequencies above 20k through different processes (downsample/LPF, whatever)...
and then maybe explain to me what is causing the difference because to me there is an obvious difference when downsampling or even just a 24db/o or more LPF @20k in the 96k file...
maybe i'm a dog...