What sample rate do you work with?

[PurpleSunray]

Throwing in some fuzzy theory... the best sample rate is the one natively supported on your audio interface (or the one of your audio engineer).
All do support 48kHz, some ($$$) do also support 96Khz.

Ok, you will argue now that re-sampling to native DSP rate is no big deal there is almost no quality loss (i.e. using a sinc interpolation), but I will argue that this is processing code which doesn't need to run if you use the native hardware rate all the way already.

Also using 96kHz as a sample rate on the project to avoid aliasing will make Intel (or any other CPU vendor) happy, but I don't think it has any effect. Anti-alias filters that need oversampling to function properly should be implemented to actually do oversampling and not rely on the user to select a high master-sample-rate. i.e. if you implement a synth that has aliasing issues on 48kHz sample rate you should implement oversamping internally to avoid that... not ignore it and rely on the user to switch the while project to 96kHz .. just to fix your ugly 48kHz-aliasing-synth *g*

[aciddose]

still though, 88200 reflects what 44.1 will sound like with greater depth - seeing as 44.1 is a norm (and has the same dillema).

What reason do you have to believe this? It is patiently incorrect.

A single 1khz sine will be represented equally at 2k, 20k, 44.1k, 48k, 88.2k, 96k or 1mhz.

[aciddose]

Anti-alias filters that need oversampling to function properly should be implemented to actually do oversampling and not rely on the user to select a high master-sample-rate.

This argument is fallacious.

Should a person floating in vacuum who requires air to breathe simply breathe air without any sort of air breathing apparatus?

Or is your argument that people who need air to breathe shouldn't exist?

I think you misunderstand the topic you're attempting to discuss. Learn a bit about a topic, then you can discuss it while maintaining some semblance of rationality.

[PurpleSunray]

Anti-alias filters that need oversampling to function properly should be implemented to actually do oversampling and not rely on the user to select a high master-sample-rate.

This argument is fallacious.

I think you misunderstand the topic you're attempting to discuss. Learn a bit about a topic, then you can discuss it while maintaining some semblance of rationality.

Happy to get some inside from you.

So where you avoid anti-aliasing by increasing the output sample rate?

I know aliasing is an issue in sampling. If they want to fix it via over-sampled anti-alias-filter they can do so... ideally doing the sampling already at high rate and then filter. The output rate of the capture device shouldn't be affected by that, i.e. changing target sample rate should not affect the oversampling on the internal anti-alias filter.. should it?
I also know about aliasing on synths, especially on the OSCs. Also here, simply run the OSC a high rate, than anti-alias on oversampled filter and afterwards convert to output rate.

Or is your argument that people who need air to breathe shouldn't exist?

No it's not.
As you say, I have no clue - but want to understand what this people are.
Anti-aliasing can't be this person unless I miss something.. if you want to avoid that simply oversample before the anti-alias-filter, no need to change input/output rate for that.
So I just don't get the point.. According to the sampling theorem you can exactly reconstruct all frequencies on a signal that are below half the sample rate. So if you go with 48kHz, you are fine until 24kHz which is way above any human ear can hear. So are we talking about creating dog-music or what is the secret behind all this which I do not understand?

There is just soooo much guessing and gut-feeling involved on on this thread.. all I'm looking for is an explanation why you need >48kHz for audio production. One I can understand (DSP / math would be nice, rather than speculation).
You seem to know what's the secret behind.. so please tell me: where you do run into anti-alias issues on 48kHz, which are gone when changing to 96kHz as a master sample rate?

[aciddose]

It doesn't make sense to say that "plugins which need anti-aliasing should use a filter that works without oversampling" because they're the same thing!

For example a distortion plug-in must over-sample the signal before it works, otherwise there is nothing to filter! Aliases appear when you sample a signal with frequency content greater than nyquist. The only way to correctly sample a signal is to increase the rate until nyquist is greater than the highest frequency component in the signal.

https://en.wikipedia.org/wiki/Aliasing

Only at that point can you apply a filter to remove the high frequency components, then you can re-sample at the lower rate.

This is why it makes the same sense as to say an astronaut should be able to breathe without a spacesuit. This is the same as saying that astronauts should not exist... astro-naughts.

[aciddose]

I think your misunderstanding is much greater though. You must do a lot of reading, read this entire thread and many more threads on the topic to begin to understand the basics. Get some books about DSP and similar. Build some of your own plugins including both synthesizers and distortion or other effects.

Then you will begin to understand the complexity involved.

[aciddose]

If I try my best to guess what you're asking, I might be able to give you this simple example:

Although please note I've already posted this same explanation several times in this thread! Read the thread!

The existing software may use a filter which results in some loss at some specific frequency. It is inevitable regardless of which filter is used that some loss will occur within the band (less than nyquist) of any sampled signal when it is applied.

If we repeat this process:
base rate -> oversampled rate -> filter -> base rate

... Many times in repetition, these effects will accumulate to unacceptable levels.

By using a higher base rate we can move the primary component of these effects out of the base rate band and apply the filter only once, after all other processing is complete.

For example:

input (base rate) -> plugin1 -> plugin2 -> plugin3 -> plugin4 -> output (base rate)

If each of these plugins must oversample and filter, the effects of the filters will be cumulative at the output base rate.

Instead, we can do the following:

input (base rate) -> oversample -> plugin1 -> plugin2 -> plugin3 -> plugin4 -> decimate + filter -> output (base rate)

Now the effects are mostly cumulative outside the base rate band and the affected portion of the output signal is filtered away by the final filter stage.

Not only that! In cases where the oversampling process applied to all the plugins satisfies the needs of the plugins, they need not further oversample the signal.

Think of this in terms of redundancy.

3x + filter ->
plugin1 requires 3x oversample.
filter + /3 ->
2x + filter ->
plugin2 requires 2x oversample.
filter + /2 ->
5x + filter ->
plugin3 requires 5x oversample.
filter + /5 ->
8x + filter ->
plugin4 requires 8x oversample.
filter + /8 ->

If we instead wrap all plugins at an oversampled rate of 8x the base rate, none of the plugins require the redundant filters!

8x + filter ->
plugin1 requires 3x oversample.
plugin2 requires 2x oversample.
plugin3 requires 5x oversample.
plugin4 requires 8x oversample.
filter + /8 ->

The overhead of processing at 8x rather than 2x, 3x or 5x may be less than the cost of processing the redundant filters, or it may be greater.

• 2x = 400% overhead
• 3x = 266% overhead
• 5x = 160% overhead

[mrw]

I would work at 88.2Khz if I had CPU that could handle that.... right now I'm at 44.1Khz

it still sounds good tho

[Shabdahbriah]

I use both 24/96 and 24/48 (w/dual recorders) for field recording/sampling. Internal (DAW) synth/effects stuff, like electronic bleeps/bloops, blips/glitches and chirps... 16/41.1.

This is one of the best articles (in addition to those by SOS) I've read regarding this topic:

http://www.trustmeimascientist.com/2013 ... n-it-isnt/

YMMV

[Musicisbest]

I use both 24/96 and 24/48 (w/dual recorders) for field recording/sampling. Internal (DAW) synth/effects stuff, like electronic bleeps/bloops, blips/glitches and chirps... 16/41.1.

This is one of the best articles (in addition to those by SOS) I've read regarding this topic:

http://www.trustmeimascientist.com/2013 ... n-it-isnt/

YMMV

Interesting, thanks for the link.