48k and 44.1k

[aciddose]

Even linear can work well as long as it is anti-aliased correctly.

Naive linear is bad as are naive splines and other simple methods. With anti-aliasing filters however they work very well.

A slightly more expensive IIR implementation using a low-pass + notch also works very well. The advantage of these methods is that they introduce only minimal ringing, while a sinc interpolation always introduces a lot.

Much like in image resampling a little bit of aliasing (otherwise inaudible) is often times better than minimal aliasing with a lot of ringing.

[pixel85]

Ps. resampling lower SR to higher is pointless. It doesn't improve anything and can even damage sound in further processing to lower SR for final product (like cd audio, mp3)

So you're saying oversampling is a disaster?

I mean resampling already recorded audio (like from 44,1 to let say 96kHz which doesnt add anything except empty bits), not oversampling effect processors

[aciddose]

They're the same thing!

The difference is with pre-recorded audio you can use much higher quality filters and do all processing ahead of time.

We know what has been recorded already, which isn't the case in real-time or when processing the output of another synthesizer or so on.

The real-time capable filters are much lower quality and do introduce the negative effects you're talking about unless you run at a higher sample-rate which moves most if not all of those negative effects out of the audible band.

Pre-processing your audio clips to re-sample at the higher rate is recommended. In fact it is by far the best option available.

The alternative is to use poor quality real-time filters which are built in to your host or other software utilizing these audio clips at a rate which differs from the real-time sample rate. This will require a significant additional amount of processing time compared to pre-processing the samples which requires zero additional real-time processing.

[camsr]

The alternative is to use poor quality real-time filters which are built in to your host or other software utilizing these audio clips at a rate which differs from the real-time sample rate. This will require a significant additional amount of processing time compared to pre-processing the samples which requires zero additional real-time processing.

But it does increase demands on memory throughput (hard drive or SSD).

[camsr]

resampling lower SR to higher is pointless. It doesn't improve anything and can even damage sound in further processing to lower SR for final product (like cd audio, mp3)

The OP stated his client wanted the format to be 48khz, so who's right?

[aciddose]

... This will require a significant additional amount of processing time compared to pre-processing the samples which requires zero additional real-time processing.

But it does increase demands on memory throughput (hard drive or SSD).

Not entirely because you would have had to perform the re-sampling anyway.

It does of course increase the bandwidth from long-term storage although in many if not most cases these days many of these files are loaded into system memory from long-term storage ahead of time.

For example a five minute recording using 32-bits per sample in stereo at 96kHz = 5 * 60 * 2 * 4 * 96000. That is ~220mb. In the typical system with 8gb of memory available (often times we'll see 32 or more these days in a custom DAW) you could fit more than 32 of these tracks in memory.

Not to mention where bandwidth matters having a modern SSD for projects you're actively working with is advisable as the bandwidth is often twice or more for a SSD.

[camsr]

For example a five minute recording using 32-bits per sample in stereo at 96kHz = 5 * 60 * 2 * 4 * 96000. That is ~220mb. In the typical system with 8gb of memory available (often times we'll see 32 or more these days in a custom DAW) you could fit more than 32 of these tracks in memory.

Not if you still use a x86 daw because you want to use old plugins
I suppose I should try bridging x86 plugins in x64 host, still haven't made the switch because of it.

[aciddose]

If you want to act like it's the 1990s ...

Actually even then you should be able to map >2gb at a time of 64gb maximum. Each individual plug-in running in 32-bit for example can have its own address space allocated which helps to avoid running out of memory.

Also most 64-bit hosts run 32-bit plug-ins fine using the same method and since the host would typically manage these tracks such a limitation doesn't apply.

[kmonkey]

If you want to act like it's the 1990s ...

When you are saying this above and this bellow:

They're the same thing!

The difference is with pre-recorded audio you can use much higher quality filters and do all processing ahead of time.

We know what has been recorded already, which isn't the case in real-time or when processing the output of another synthesizer or so on.

Pre-processing your audio clips to re-sample at the higher rate is recommended. In fact it is by far the best option available.

The alternative is to use poor quality real-time filters which are built in to your host or other software utilizing these audio clips at a rate which differs from the real-time sample rate. This will require a significant additional amount of processing time compared to pre-processing the samples which requires zero additional real-time processing.

It almost looks like music was not existed in era when people could not afford expensive computers, computing and storage

I do get your contribution and quite frankly even the most partizan can not argue with you because technically and theoretically you are right by all means.

However...

To my amazement even in pre 1990 era people was able to create amazing music with supposedly shitty 16-bit reverbs, 8-bit samplers and all kind of crazy stuff which is by KVR standards outdated

Even at their shitty computers without antialiased Jesus Budda Christ!!!

Not only these same people created music genres which we are following today (in a bad way), but for some reason you will hear even at this forum that music had more life before then today.

As i see it, new people with modern computers should follow your guidelines if they are strictly ITB. Sure it is a waste of resources at some point (really depend on their plugins in their chain), but in the long run working at 96khz should provide them better results with less possible errors - but again this depend on what/which plugins they use.

For OTB well like i said. People are doing amazing music even at 44khz / 48khz.

I guess my point was that what you are saying is hard to beat, and working ITB on modern computer this is no question, but sometimes this does not apply to all workflows. I guess OP should just start working 96khz if he is ITB and on modern computer.

[aciddose]

No idea what you're talking about. I'm not aware of any mainstream music that was produced via PC during the 1990s.

The wide mainstream application of CG content in music only really took off around 2010. The amateur market of course moved to PC sooner, although the professional market still maintained revenue for studios. Now outside the live/performance areas there is very little use of large studios and the hardware they provide.

I would argue this is in fact entirely because the amount of memory and processing power increased radically around this time which allowed music production to become more serious. In fact it is entirely due to being able to stream 100s of tracks combined with high quality anti-aliasing filters that so much music is produced "in the box" today.

Or you can jump on your 386 and prove me wrong.

[chk071]

Regardless of the sometimes a bit heated tone of the discussion, this is a very interesting topic.

[Soundplex]

The problem with topics in which aciddose posts, is that they quickly run into discussions about irrelevant trivial matters. Nitpicking just to defend a viewpoint & to be absolutely right, is unfortunately his manner...

There's one simple way to check does higher sample rate is better or not: test it.

I agree. Just bounce your project in 44.1/48/... and compare the results. Use what sounds better and what your machine can process without problems regarding resources. Btw keep in mind there are also plugins which have a maximum processable samplerate (some of them internally work with 44.1).

[aciddose]

Speaking of "irrelevant trivial matters": the issue isn't what "sounds good" subjectively.

It's that you need to use 48000 Hz to produce a soundtrack for a modern film. There isn't much harm in re-sampling 44100 to 48000 (despite what some may tell you.)

The topic of the thread is:

hi, I normally produce music at 44.1k - but I've been asked to provide some at 48k. Is it as simple as changing the sample rate on my soundcard (which has a number of options including 48k) and in my export function on my DAW?

If so - is there any reason why I shouldn't leave it at the higher sample rate permanently for all future projects?

We've discussed some possible drawbacks (in theory) although the only issue I can see ever realistically causing trouble for anyone is:

- Running at 48000 Hz is potentially 9% more expensive.

"Sounds good" doesn't enter into this because it isn't a "sounds good" issue, it's a "the producer requires 48000 Hz material" issue.

[llze]

If so - is there any reason why I shouldn't leave it at the higher sample rate permanently for all future projects?

the question "is there any reason.." opens the discussion about sound quality.

[aciddose]

No, he wasn't asking whether he should make a subjective decision based upon sound quality. He was asking for objective facts to be presented.

Trying to make this subjective is simply a way for you to avoid having to be technical due to your ineptitude. ("You" used abstractly. Not you-you, but "a person".) If you want to make a statement of fact by all means please do.

I think we've covered all the drawbacks and advantages. If you have some I'm unaware of I'd love to learn about them.