It's simple, render at 32 for processing purposes. Then for you final production files, 24 or 16 bit when processing is completed.   For home mastering, you should even up sample your bandwidth to even 96/24 .  Many mastering engineers will up sample your tracks anyway before they start their process

BenMMusTech

drewfx1
What people who don't understand technical details don't understand is that you can't increase resolution by changing something that isn't limiting resolution in the first place.
 

I can read the technical stuff Drew - I had to to pass my first degree, and what you're missing is the notion of rounding errors.

No Ben.
 
What you're missing is the notion of how errors do and don't accumulate. And how errors buried tens of dB under the noise floor can't just magically rise above it.
 
What bit depth is needed for a given process is based on the noise floor and the number of calculations being done on the same data. Simple as that.
 
Except it's not simple because most people tend not to know either of those things for the perfectly understandable reason that they aren't readily available. One can get the noise floor from not entirely trivial measurements, but the only way to understand the calculations is unfortunately to have some knowledge of DSP programming.

We are talking here about technical issues, and reasonably well settled ones at that.

All engineering endeavors (as opposed to actually composing/arranging/orchestrating/playing) are an exercise in optimization. All engineering endeavors, including audio engineering. In order to optimize you need to know the limits of all the elements of a system, in our case the listener, the listening environment, the signal sources, and the processing engine, which includes any code used to mix and/or render.

All before we take into account personal taste.

For recording a live source you simply don't need more than 24 bits because, as stated earlier, the A/D converter can't match that. For the same reason fixed point is more than adequate. As stated previously, you can never increase the resolution in terms of frequency response or noise floor, what you record is what you get.
 
Sample rate is slightly more complex! If you are in a really quiet room with really good microphones, really good preamplifers, NO analog processing (unless you want to capture all the artifacts, which by the way will limit the dynamic range and pass band further) and a great instrument (great player doesn't hurt either) - if all of that is true then I'd probably record at 96 kHz. Otherwise 48 kHz is more than adequate.

I'd record at 96 kHz in order to preserve - in the recording - as much of the original signal as possible, or more even<G>! Keeping things as good as you can until the final mix is a great idea, and no, I don't think it is lost on us old folks (get off my lawn). Most of us "grew up" in a time when even 96 dB S/N ratio was unobtainable, and unnecessary since neither FM  radio nor vinyl discs could reproduce it<G>!

Fortunately (for me) my recording space is awful - ok, that's not entirely fortunate, but it does mean I have no excuse for recording at anything more than 48 kHz/24 bit fixed point. Which means that I don't need to consider processor power or disk space (I suppose I could probably get away with 44.1/16, but even I won't go that far!)

Where things get more dicey - a lot more dicey in some cases - is when we start manipulating the audio data. All processing in the digital domain is nothing more than math and the limits of precision and accuracy are well understood.

But if you have the horsepower why not work at greater word length? You only lose disk space, and maybe processing power in extreme cases. You won't hurt the audio, but you won't improve the source, and that is important.

For my own case, I can hear the difference between different sample rates and word lengths for some plugins (most notably some of the UA stuff). Well, I can hear the difference in a proper studio, in my studio these things have no impact. So that's another issue for optimization.

With respect, without something to compare I really can't comment on the tracks at 1331.space with respect to this discussion. Personally I found many of the tracks to be a bit too busy for my tastes, and in some cases perhaps over processed (this coming from me is almost comical!).

TL;DR
All this to say, if you are able to work with at least one order of magnitude greater resolution than you need you will be just fine.

drewfx1

Anderton
I think the rules are different for what happens outside the box compared to what happens inside the box. If you record an acoustic guitar at 44.1 kHz with 24 bits of resolution and play it back, then it will be 24 bits no matter what. 

 
Technically speaking, though the recording itself will be 24 bits, the actual resolution will depend on the playback level and the noise from all sources present - including the noise in the signal picked up by the mic when the recording was made. The limits of hearing could also apply if they weren't buried under the noise floor from everything else. IOW 24 bit is the theoretical limit of that particular digital format, not the real world resolution (which is always far less than 24 bits).

 
I know that, but those 24 bits still contain information, even if it's just noise or zeroes. I think the world has standardized on saying converters have 24-bit resolution if technically, they're capable of it because there's no way of knowing the extent to which that resolution will be compromised in a final product. So I would differentiate between useable and theoretical resolution.
 

The problem for us here is that one needs to know some complicated DSP stuff to know when a given bit depth/sample rate might be really significant, completely irrelevant or somewhere in between.

 
I find the really significant ones are audible and obvious. For example, rendering a non-oversampling virtual instrument pulse wave at 44.1 or 192 kHz is such a huge difference you might as well be listening to a different patch altogether. I did a test file and sent it around to some people with whom I'd discussed high sample rates at some recent conferences. They were truly blown away at the difference, it's not even remotely subtle. However most of the time it doesn't matter. With one amp sim, rendering at 96 kHz didn't change the tone because the amps were oversampled, but the reverb's imaging was different compared to 44.1 kHz. Go figure.
 
I think there's still a lot we don't know about digital audio and human perception. I don't play the "golden ears" thing much but I will say that back in the early days, when my converters went from 16 bits to 20 bits with the Ensoniq PARIS system, the difference was obvious and I could really hear the difference between 16 bits of real resolution (courtesy of 20-bit operation) and the 12-14 bits of real-world resolution the 16-bit converters delivered.
 
My basic guideline is if I can hear a difference, I do it and if I don't hear a difference, I don't. That's why even though I'm convinced beyond a shadow of a doubt that rendering at 192 kHz can deliver superior virtual instrument sound quality, I still record at 44.1 kHz. Why? Because there are very few times that rendering at 192 kHz does make a difference, and I can always export the file, render at 192 kHz, then bring it back into the 44.1 kHz project to obtain the benefits of 192 kHz rendering in a 44.1 kHz project. And of course, CbB's upsampling is all you need in most cases anyway.

wst3
But if you have the horsepower why not work at greater word length? You only lose disk space, and maybe processing power in extreme cases. You won't hurt the audio, but you won't improve the source, and that is important.

I agree completely.
 
The problem is that some people will claim they can hear a difference regardless of how impossible that is.  Which, since almost no one does careful measurements and/or carefully controlled listening tests, will cause others to believe they can hear it too when doing casual listening comparisons.
 
Remember a while back when there was a bug with the 64 bit engine and people were freaking out during the time before it was patched? Over errors where very few of them even make it into 24 bit output, much less approach being audible? 
 
 
And is 64 bit really good enough? How do we know it doesn't sound "smoother" if we do all of our calculations at 128 bit? And 512 bit has to be better than that, right? I mean it has more resolution and the rounding errors are lower. Who could argue that?
 
 
IOW, doesn't at some point someone who understands the relative importance of things have to point out that beyond a certain point things don't matter, and when that point has been reached?

Anderton

drewfx1

Anderton
I think the rules are different for what happens outside the box compared to what happens inside the box. If you record an acoustic guitar at 44.1 kHz with 24 bits of resolution and play it back, then it will be 24 bits no matter what. 

 
Technically speaking, though the recording itself will be 24 bits, the actual resolution will depend on the playback level and the noise from all sources present - including the noise in the signal picked up by the mic when the recording was made. The limits of hearing could also apply if they weren't buried under the noise floor from everything else. IOW 24 bit is the theoretical limit of that particular digital format, not the real world resolution (which is always far less than 24 bits).

 
I know that, but those 24 bits still contain information, even if it's just noise or zeroes. I think the world has standardized on saying converters have 24-bit resolution if technically, they're capable of it because there's no way of knowing the extent to which that resolution will be compromised in a final product. So I would differentiate between useable and theoretical resolution.
 

 
Yeah, I was more just trying to clarify something to make sure people didn't misconstrue what you were saying.
 

 

The problem for us here is that one needs to know some complicated DSP stuff to know when a given bit depth/sample rate might be really significant, completely irrelevant or somewhere in between.

 
I find the really significant ones are audible and obvious. For example, rendering a non-oversampling virtual instrument pulse wave at 44.1 or 192 kHz is such a huge difference you might as well be listening to a different patch altogether.

 
And that's a good example of something that can be unambiguously audible to the point that you don't have to bother with careful blind testing. And I'm betting that if I ask, what you'll describe hearing fits pretty much exactly with what we would expect you to hear. I'll just say that for some things what people describe "hearing" doesn't always match the artifact they claim to be hearing. Hmmm....
 
I should also add that even in this very good example of the benefits of oversampling, if the virtual instrument had been coded to use what's known as "band limited waveforms" then the oversampling might not have been of any value.
 
The point being that it's hard to come up with simple guidelines.
 
Having said that....
 

My basic guideline is if I can hear a difference, I do it and if I don't hear a difference, I don't.

That's a pretty good guideline. 
 
But I would suggest we just have to be careful when we start getting to the point where differences are harder to hear and/or can only be described with vague subjective language that even people of your level wouldn't be quite sure how to measure if we wanted to. Or one can do a careful double blind listening test, because unfortunately we are all basically programmed to sometimes imagine we hear stuff that isn't there. And as the saying goes, this applies to everyone, including the people who don't think it applies to them.

it's quite possible that the higher sampling frequencies are delivering superior sound because the reconstruction filters are performing more effectively at the higher sampling frequency. possibly, the interpolation being performed between the discrete samples is more pleasing to the ear with higher frequency sampling. 
 
Nyquist-Shannon sampling theorem does not provide a simple, straight forward way to determine the correct minimum sample rate for a system because it requires that the signal be perfectly bandlimited and we cannot do that

rabeach
Nyquist-Shannon sampling theorem does not provide a simple, straight forward way to determine the correct minimum sample rate for a system because it requires that the signal be perfectly bandlimited and we cannot do that

We don't live in a perfect world so we don't need a perfect filter.
 
I would say you can determine it by starting with the following questions and then using a good filter design algorithm:
 
1. What frequency range are we interested in (i.e. what is the pass band)?
 
2. How much stop band attenuation do we need?
 
3. How flat does the pass band need to be?
 
4. How much latency/CPU can we afford?
 
Doesn't that, and maybe a few other details pretty much spell it out?

rabeach
it's quite possible that the higher sampling frequencies are delivering superior sound because the reconstruction filters are performing more effectively at the higher sampling frequency. possibly, the interpolation being performed between the discrete samples is more pleasing to the ear with higher frequency sampling. 
 
Nyquist-Shannon sampling theorem does not provide a simple, straight forward way to determine the correct minimum sample rate for a system because it requires that the signal be perfectly bandlimited and we cannot do that

You don't need a perfect filter, just one good enough to satisfy the parameters of the human auditory system. IOW, one that can attenuate unwanted frequencies to the point of inaudibility. The filters in modern converters are more than adequate. Remember, they are no longer attempting to chop off frequencies in the kilohertz range, but in the megahertz range - a much easier task that doesn't require sharp slopes or great precision.
 
Fortunately, interpolation doesn't enter into it. As long as any missing frequencies are outside the range of human hearing, they're not going to be missed.
 
You can prove that to yourself experimentally. Take a file recorded at 96 KHz, preferably one with a lot of high-frequency content, and load it into Cakewalk. Apply a low-pass filter set to 30 KHz or higher (you'll need a high-end equalizer; Pro-Q2, for example, can do 30 Khz). Insert an automation envelope to adjust the cutoff frequency of the HPF and throw in some wildly varying nodes such that the filter wanders from 30 KHz to 20 KHz with random stops in between. The reason you use automation is so you can listen blind, without knowing where the filter's cutoff is at any point in time. Trust me, you will not hear the filter doing anything at all.

bitflipper
 
Fortunately, interpolation doesn't enter into it. As long as any missing frequencies are outside the range of human hearing, they're not going to be missed.
 

I think he meant the interpolation done by the reconstruction filter. Or perhaps by the, um, interpolation filter in an upsampled DAC.