Project Sample Rate Reduction Test

I was working on a project with lots of VSTi’s and plugins and the old VST meter was starting to ping. For interest sake I decided to save the project under another name and reduce the sample rate in the new project from 96 to 44.1 Cubase went away and did its thing (quite clever actually – did all the SRCs to the audio files and prompted to save the original audio files and so on…). Once finished I played the tune and noticed something strange with the drums. I have a couple of instances of Groove Agent in the project and when I solo’d them I noticed they were all distorted. In fact, it appears that Groove Agent did not get the message about the sample rate conversion and continued to render at the higher sample rate (or so it seems as I cannot really confirm this). In any event Cubase did a pretty good job with overall process except for Groove Agent.

One other thing… I was reading about the seemingly endless debate about sample rates – those who assert there is no audible difference between 44.1 and anything higher (be it 96 or 192 or whatever). At my age I cannot claim ‘golden ear’ status but I have to say that even my grizzled tympanic membranes could hear a significant difference between the 44.1 project and the 96 project. Who knows, maybe there were other conversion issues of which I remain unaware but even after fixing the drums the 44.1 version sounded tinnier than the 96 version.

On the upside it was great for VST load… Dropped it by about 40%.

Aloha c.

Nice report and info.
Thanks for that.

I too can not so much ‘hear’ as ‘feel’ a difference.
But not every time and not with certain types of music.

Yes, that’s a much better way of putting it… I only tried this on one project… I think I’ll give it a go on something else and see what happens… Thanks! C

Coming from 44.1 and 48 all of my recording life to 88.2khz recently… I will never ever go back to anything lower.
It just doesn’t “feel” more accurate when I mix but it sounds significantly better.

I’m one of the guys who can’t hear a significant difference. Working with lots of audio, rarely VSTi which are told to sound better at higher rates.

For me it’s not worth the hassle of significantly increased system load. But this comes up every now and then, think I’ll try again @ 88.2 :wink:

I guess it depends on your system. With my specs and usage, I have not experienced any increased system load issues. The only difference is the quality of the sound.

I wish Cubase would be able to downsample on the fly, so we could record at 192 kHz but put only 48 kHz of load on our CPUs.

This would also be amazing for the ASIO load, because we could run everything at 192 kHz, but only every fourth sample is really processed while doing “online” work.

So we could run comfortably at buffer sizes of 512, while maintaining “almost zero latency” feeling.

Sadly, no one @ Steinberg cares for this simple, straightforward idea.

Ok, it’s a while ago (one computer generation back) that I’ve tried last. Mixing a lot with UAD plugins, higher rates cut down instance counts on the plugins that do not upsample anyway. That was my most serious problem. And maybe the reason for not really recognizing a quality shift when a certain amount of upsampling happens anyway :question: Not sure…

The less the content changes, the less noticeable will be sample rate differences.

The Nyquist Theorem says that given an INFINITELY repeated waveform, the maximum possible frequency is half the sample rate.

That means that synthesized music that has only very slow changes will likely be better tolerated at low samples rates than sparse vocals and acoustic instruments with rich harmonics and rapidly varying sounds.

As many have quoted here over the years, tests have shown that most listeners would be hard-pressed to tell the difference between samples rates, presumably after only one conversion between the two. So there may not be much point in making your music available in anything greater than CD quality (44.1kHz 16-bit).

However, processing at different sample rates throughout a whole project is another matter, as there is lots of processing going on. That is why Universal Audio internally upscale to 192k on most of the UAD plugins to ensure the accuracy of their emulations. It is engineering common sense, as one cannot expect to have an accurate output if performing thousands of processing steps at the same accuracy as expected at the end.

With our music, we are mainly recording sparse vocal and acoustic instruments using condenser microphones, which pick up EVERY little sound. I record at 192k, so I can clean up the best with RX3. It also allows the reverbs to create the vocal richness (UAD EMT140 plate) and realistic room sound (REVerence LA Studio).

Of course, higher rates require more system resources, which is giving me aggravation when mixing down on my current system. The solution for me is only upgrade!

Also, if using samples, most of which are at 44.1, they will be up-sampled on the fly EVERY time, using more CPU resources. So it is best to freeze or bounce to audio once an instrument part is not likely to change.

This makes perfect sense. I would assume that lower sample rates within a given project multiplied by the number of various DSP instances can cause geometric colouration (or degradation) as ‘errors’ multiply by ‘errors’ (meaning ‘inaccuracy’ for a given operation is inherited by the next operation and so on). In any event I certainly prefer the overall experience at higher sample rates. As to what happens once the music is finished and put into the world I cannot say as I think I am my only listener :slight_smile:

Think of a 5K sine wave. A 44k sample rate can only detect its phase to within 5000 / 44100 * 360 = 41o. What happens when the note that that 5K is part of is being played on guitar with some vibrato and dying down. How accurately will the phase be represented then?

The problem is that there seems to be no definitive answer for all this, so I err on the side of engineering conservativeness and tend towards over-engineering. That Universal Audio has bet their reputation on it without advertising it, tends to back up my decision.

As stated above, I can’t hear a significant difference. Never did extensive work at higher rates above 44/48 though, so I’m unexperienced with this.

But well, it’s music we’re talking about. Not everything might happen in a ‘measurable dimension’ (that’s where we might take some pleasure from, otherwise we’d perhaps all be bankers, architects, scientists…).
I don’t think the better ‘feel’ someone might have at high sample rates is definately ‘not there’ nor would I say it is. All of that belongs to indiviual psychological aspects, a great confusion of personal experience (that might differ from day to day), marketing blurb (higher numbers better in this case) and the wish to use the best available technique. Add to the list what comes to your mind.

It’s still about the content. The content is about the creators of that content (spirit, skills). Add the potential listener into this equation and you have a mass/mess of factors having an impact on how the music is finally perceived. There’s no control and that’s cool.

Fake philosophies aside, breaking it down to what works best for you to get the thing done is simple and pragmatic. 44.1 seems to be alright for me still :sunglasses: If you want more and have the power, why not.

Some points:

  • Response outside one’s hearing range is irrelevant. Banging on about it is a red herring. It is the real in-band accuracy that I am interested in, not what some ‘golden ears’ 20yo can hear. You brought this aspect up, so stop wasting your own time.
  • One conversion does not a song destroy! But 10,000,000 calculations can.
  • Bob Katz also states in his book that digital filter design seems to be the first thing to suffer in cheaper equipment. But his is not really the key real issue for DAW users. It is the effect of lots of processing beyond a single conversion.

Of course, the actual calculations on the digital side are far more involved than any simple explanations can cover.
I have also covered much of this on the Soundsonline forums, and it was quite clear from those that there are a lot of misconceptions, especially about the Nyquist Theorem.

What I find most interesting is that the human body uses a more digital system than we use in our DAWs, and that we are only trying to fly below the body’s reality threshold.

First of all, everything above 48 kHz in terms of sample rate is esoteric.

Modern filters are phase linear even when using a very, very steep slope, so there should not be any problem with the really relevant cutoff frequency of about 16 kHz (kids don’t really appreciate music anyway, their hearing abilities are not relevant to us).

Since Nyquist we KNOW even a formula to find out about the sample rate we need. And we also know, from Fourier, that even if there WAS stuff happening in addition to Nyquist, it wouldn’t matter, because all this stuff would be above what we could hear.

In my honest opinion, anyone trying to sell you anything above > 48 kHz (EXCEPT for processing, like oversampling, but this has other reasons, which are purely mathematical and are therefore justified) for playback or recording is selling you snake oil and is wasting your CPU power and your harddisk space.

96 kHz is the same to me as those “EUR 3.500,- HDMI cables”.

So, my home studios format is 48 kHz (because I use optical interfacing) and, of course, 24 bits. It’s just not possible to get any better, biology and mathematics are putting up a limit here.

If you keep trying to make sample rate discussions about upper hearing limits rather than accuracy of IN-BAND (that is, frequencies we CAN hear) processing, you are just trying to spread FUD. It is already well known what the limits of human hearing are, but that is not what really needs to be quantified regarding sample rates.

If you have no better arguments than ‘I can’t hear it, so it is a waste of time for anyone else’, or just being dismissive, don’t bother vainly trying to contribute any further to this discussion, because you are not clarifying the main issue regarding sample rates for recording and processing BEFORE making the final down-sample for normal listening.

Just to summarise some key points:

  1. Engineering common sense requires that the precision of what you start with, and process it at, is higher than the required end result. Higher bit depth gives greater precision in the amplitude domain, but nothing greater in the time domain, where phase and level changes occur.
    The more complex the total processing, the greater the precision required.

  2. The Nyquist Theorem is for the ONE special case of an infinite repetitive waveform, and says nothing about how much is needed to cover real world situations. Unfortunately, too many assume it does and stop exploring its limitations, which makes it difficult to arrive at minimum recommended sample rates for different types of music, or even a definitive one-size-fits-all. A top-level studio can just afford to have the best, but for us lesser mortals, some informed guidance may help our pockets cover all our needs better.

    Supporting higher sample rates are:

  3. Universal Audio internally uses 192k on most of their UAD plugins. This is only mentioned in their manual and nowhere else, not even in their advertising. It does reduce the capacity of their UAD hardware, so it has a significant down-side, but they still use it. I deduce from that, that it IS REQUIRED for them to be able to accurately emulate REAL-WORLD devices.
    It is this that originally piqued my interest in higher sample rates. Companies generally don’t saddle their products with performance limitations unless there are demonstrable benefits. Yes, they may sell more hardware by artificially limiting capabilities, but they advertise on the basis of the fantastic number of plugins one can use for given hardware. They could have just said it is all magic, but they have tables in their manual of all the extra latency required for each plugin, purely because of using 192k, which no-one in their right mind would want to saddle themselves or their users with if they had a choice. The growing competition has also not resulted in them abandoning this.

  4. Bob Katz, in Paul Gilreath’s The Guide to MIDI Orchestration, recommends recording, or at least up-scaling on first use, to 96k, and only down-sampling and dithering at the very end.
    In his book, Mastering Audio, he cites other peoples’ research into advantages of higher sample rates. In the book, he states that he has designed digital filters that work very well at 44.1kHz, yet he still recommends using higher sample rates.

    I would truly like to have some KPIs by which I could measure our recordings and have a formula by which to calculate the minimum suitable processing sample rate that would cover our target audience, INCLUDING those who appreciate quality.

…and I would like to see double blind testing, performed by those self proclaimed “audiophiles” (including those “experts” who seriously think that there is a difference between digital cables… :laughing: ) to see if there is any (!) perceivable difference between 48 kHz and 192 kHz (to take the two most extreme examples).

I would BET that, as well as with and without heavy processing, there is none. This is snake oil, there is no scientific basis for any other result, because the difference in the information between the “low” and the “high” sample rate audio signal is not audible to the human ear.

All you add is ultrasound, which is such a waste.

Again, I agree that processing must be done (in some cases) using oversampling, possibly heavy ultrasampling, but this is for mathematical reasons during the process (FM synthesizers being a good example), but after all mathematical transformations, the result can be easily scaled back to the regular sampling frequency.

It’s a bit like electrial engineers calculating with the square root of “-1”, which is “i”, but the final result for the end user doesn’t contain any complex numbers.

I stand by what I said: 48 kHz of sample rate is all you need, except if you are producing for bats. :stuck_out_tongue:

So all you have to offer is assertion, ridicule and your unproven reputation? Wow, what a tosser!

No, I have science on my side.

I never claimed any reputation, all I used here was scientific facts. And it IS a fact that the Nyquist theorem requires only 2x 16 kHz = 32 kHz of sampling rate for the human ear, but to be able to use the necessary low pass filters, using 44.1 kHz or 48 kHz is the right thing to do.

Granted, very small children can hear beyond 16 kHz, but they are not relevant to us, because they a) can’t tell anyway and b) are outgrowing this very soon and c) that still doesn’t require 96 kHz or even 192 kHz of sampling, because even a kid which just was given birth to can’t possibly hear beyond, say, 25 kHz.

No snake oil for me, I prefer hard facts, not some esoteric babble.

In addition, I totally agree with Bredo. The filter design and implementation is the most important aspect of all of this, nothing else, because the ADC converters themselves are amazingly good nowadays.

The Nyquist Theorem proves NOTHING about ANY signal that is NOT (INFINITE && REPETATIVE).

I am not into such very particular sounds, so the Nyquist Theorem does not help me, other than for worst case, but not best practice.

It is you that is making some wrong assumptions about the veracity and applicability of your ‘facts’.

Do you understand Fouriers theorem, too?

Of course, using Nyquist alone, one could still claim “yes, but there are other things happen, which can only be captured using higher sampling frequencies”. I fell for this misconception, too, a few years ago.

Then some mathematician friend of mine explained Fouriers theorem to me and made it clear to me that anything, that would not be captured by a normal sampling rate, which would not be ultrasound.

There simpy IS only ultrasound you add when you increase the sample rate beyond a certain point… except, and I totally give you THAT, phase nonlinearities caused by suboptimal low pass filter designs, but they are not an issue anymore nowadays.

And yes, I’m all for 192 kHz processing in some plugins, processors, also synthesizers. But neither for recording, nor playback.