Still waiting to hear if Prism and Wavestation have a HQ/oversampling mode and if there is a difference.

Good point, I checked. In stand alone mode both of these plugins have the option to set the sample rate to 96 K. Once you do this though the session sample rate will be asked to change over to match.
 
When running as VST's inside a host program such as Studio One it seems the sample rate is locked to the host session and cannot be changed.  If you create a seesion at 96KHz though the sameple rate inside the plugin switches accordingly.
 
Update
Just completed second experiment. This time I kept the synth component constant but only changed the session sample rate.
 
https://www.hightail.com/download/ZUcxSlJ5VnNoMlVVV01UQw
 
I created a 5 part midi multi-timbral sequence for the (hardware) Kawai K5000 just utilising on board sounds. More rhythmical groove this time but still lots of additve patches rich in high end harmonics. Coming in through the same sound card. (EMU 121M Analog ins and outs. Same converters as Pro Tools HD interfaces. Can go up to 192Khz)
 
The good news is that the multi timbral sequence plays the same in both cases. I just printed in stereo the output direct from the K5000 itself into the analog input of the sound card. Twice, at 96 Khz session and 44.1K session. I then exported the 96K stereo print downsampled and out to a 44.1 K rendered file. All these files are at 24 bit. Timing and level should be pretty accurate.
 
These two files now sound very similar, even though they both recorded complex additive sounds. I would still not mind people's opinion as to how the 96K rendered down to 44.1 version sounds compared to the straight out 44.1 K version.
 
Something interesting here. The waveforms of both of these waves do not match. Also I cannot get any sort of a null no matter what I attempt to do. The two waves are too different to allow for any usable null. Despite the music sounding so similar in both cases.
 
Thanks to Craig for starting this thread. It was an interesting experiment for me. It really showed up the differences when these things are running as virtual instruments are beginning life at 96 kHz. Perhaps after you are back in the forum land of the living Craig you could check out the second pair of waves and I would be interested in your take on the differences or lack there of etc..
 
In fact this thread in a way has changed my thinking. To perhaps building up a system that is totally running in 96 kHz and 24 bit all the way from end to end. And as I like digital mixers the obvious companion to something like this is Yamaha 02R96V2 or the DM2000VCM with its 8 x 96K effects processors on board. Serious or what. Computing wise I think something like Studio One running on the Apple Mac Pro in tandem with a UAD Apollo 16 interface would be pretty sweet too.

Time to bump this thread, it is getting lost and is seriously important!
 
Craig has brought up a very valid point. I think my second experiment confirms the fact that when you are recording from the outside into your DAW and even when that instrument is something very complex sounding such as a Kawai K5000 it can be said there is probably very little difference between sessions running at 44.1K or 96K. Like we used to think and that seems to hold true.
 
But NOT when a VST synth such as Native 'Prism' is being used to create the parts and being used digitally right from the get go. Then it seems 96K makes a big difference. So much so that I was thinking of doing a rebuild for a new setup but now I am re thinking going 96K all the way as it makes such a difference to some VST's running at that sample rate.
 
Something to think about for sure.
 
 

Jeff Evans
Craig has brought up a very valid point. I think my second experiment confirms the fact that when you are recording from the outside into your DAW and even when that instrument is something very complex sounding such as a Kawai K5000 it can be said there is probably very little difference between sessions running at 44.1K or 96K. Like we used to think and that seems to hold true.
 
But NOT when a VST synth such as Native 'Prism' is being used to create the parts and being used digitally right from the get go. Then it seems 96K makes a big difference. So much so that I was thinking of doing a rebuild for a new setup but now I am re thinking going 96K all the way as it makes such a difference to some VST's running at that sample rate.

 
Thank you very much Jeff for chiming in with your findings. You totally understand my original point and I'm glad it's been an interesting ride for you. The irony, of course, is my doing this experiment in the first place to show that 96kHz didn't make a difference. Live and learn
 
I just landed in New York and will be on the first panel tomorrow at the New Music Seminar discussing all this. I'll review the previous posts to see if there's anything that requires comment, although it seems like you've taken care of that.
 
I've also found out a few more points of interest and will add those in when I get a chance. BTW Jerry Harrison (Talking Heads) will be on the panel and we've known each other for quite some time, so I'll get his comments on this as well. Should be interesting!

bitflipper

According to IK Multimedia's chief engineer, physical guitar amps generate harmonics well above the audible range and part of their emulation process is to reproduce those frequencies. He also said that high-gain amp sims often deliver 60dB of gain.

 
Amplifiers, yes. Guitar speakers, no. And most microphones couldn't pick them up anyway.
 
Stick a microphone in front of a guitar speaker cabinet and play a fat distorted chord, record it at 192 KHz and analyze the spectrum. The amplitudes of supersonic harmonics will be very, very small - if detectable at all - and they'll be the product of unpleasant intermodulation distortion. Your typical guitar amp and speaker will roll off steeply over about 12 KHz, and even if it didn't your microphone won't pick up much beyond 20 KHz. Certainly not the ubiquitous SM-58 that's so commonly used for this purpose.

I think you're still missing the point, it's not about high frequencies we hear, it's about high frequencies that cause distortion in such a manner that it creates artifacts at frequencies we can hear...and which speakers can reproduce, and microphones can pick up.
 
The thing that surprised me the most about the experiments I did wasn't that the non-oversampled processors and instruments sounded better at 96kHz, but that the audible improvements remained when downsampled back to 44.1kHz. In retrospect that probably shouldn't have been a surprise, because 44.1kHz is quite capable of reproducing sound within the audible range. So if I could hear an improvement in the audible range at 96kHz, it makes sense it would translate to 44.1kHz.

Thanks Craig for coming back! I was beginning to think no one was going to say anything on this again!
 
I think the fact that my second experiment shows even when recording a complex sound in through the sound card the differences between 96K and 44.1K sessions are much leas obvious. And this is how many Sonar users for example will be using this approach producing non synthesised music doing bands or their own music and stuff etc..
 
But what about those of use who want to use a digital VST such as Prism and use it to create music. Then this is where the 44.1 vs 96K thing comes into its own. It is just seriously different and pretty obvious to me. This is obviously a situation where the higher sample rate is way preferred and just sounds much nicer. The fact that it translates down too does not surprise me either.
 
I use a lot of hardware synths too but I am also a big fan of many virtual instruments. I can just imagine how many others will sound nicer at 96K. The reason I am re thinking a complete setup at 96K all the way through from end to end. For me it is the only way to go now.
 
And when that system is being used to just record more normal things (eg a band) I bet it will still sound a little nicer in the long run too. And that nice sound will be translated down to 44.1K 16 bit and still be there.
 
The reason things can sound so cool at 44.1K and 16 bit is shown in this experiment. (Sorry for those of you that have heard about this.) If you take a serious analog signal (finest turntable, pickup, RIAA eq etc and a Sheffield Lab record!!!) and feed that to one side of an AB switch. Now the analog signal is also fed through a A to D and D to A all at 44.1K and 16 bit and feed to the other side of the switch. This has been done in a room full of experts with amazing gear under almost ideal conditions and mostly NO-ONE was able to reliably pick the analog signal every time. Interesting don't you think.
 
What this says is that if a signal sounds wonderful to begin with, then running it down to 44.1K and 16 bit has no bearing on that wonderful sound.  Think of Prism running at 96K as a wonderful sounding signal to begin with. The fact that it begins as a digital signal is not so much of consequence.

 I think you're still missing the point, it's not about high frequencies we hear, it's about high frequencies that cause distortion in such a manner that it creates artifacts at frequencies we can hear...and which speakers can reproduce, and microphones can pick up.

No, I'm not missing the point. This angle has been discussed many times before.
 
Supersonic frequencies do mix acoustically to produce sum and difference frequencies, and the latter are indeed audible. Cymbals, for example, can have twice as much energy above 20 KHz as below it. That beautiful mash-up of frequencies bounce around the room and come back to the ear as a very complex - and definitely audible - sound. Without those inaudible frequencies, the cymbal would sound thin and cheap.
 
However, it isn't necessary to record those supersonic frequencies to get a great cymbal sound, because even though they're actively involved in the final sound, that activity takes place in the air, before getting to the microphone. By the time we encode it as digital data, the magic's already happened.
 
Guitar amplifiers are different from acoustical instruments such as cymbals, though, because their acoustical output is generated by components that are physically incapable of reproducing supersonic frequencies. You may very well have harmonics generated by vacuum tubes as high as 100 KHz, but most don't make it past the output transformer and the ones that do certainly don't make it past the speakers. Nobody bothers putting Earthworks microphones on a Fender Twin, for good reason.
 
Your best candidate for justifying higher sample rates probably lies in virtual instruments and effects that can create frequencies beyond Nyquist: software synthesizers, distortion plugins and fast-attack dynamics processors. None of this applies to sampled instruments, nor to non-distorting processors such as reverbs and equalizers.
 
Even in those cases where a processor or an oscillator might generate supersonic content, a mere doubling of the sample rate isn't an effective way to deal with it. While there will be fewer frequencies to alias, there'll still be plenty of potentially harmful harmonics to make trouble. You'd really need to at least quadruple the sample rate to be effective, but nobody's making a case for 192 KHz sample rates.
 
I'm not saying that there necessarily isn't some kind of justification for 96 KHz. But over the years people have tried and tried and grasped at increasingly unlikely straws to make the case and so far nobody's come up with a rationale that's supported by science.
 
So Craig, who's going to be on that panel with you? Any real engineers?

You're still missing the point about guitar amplifiers. Perhaps I didn't explain in sufficient detail.
 
Guitar amplifiers generate supersonic signals due to distortion. The levels of these supersonic signals can be significant owing to the high gains inherently used in creating distortion.
 
IK (and possibly others, but IK was willing to go on the record) emulates this characteristic of guitar amps because, simply stated, they want to emulate a guitar amp's characteristics as closely as possible. 
 
With a physical guitar amp, let's assume for the sake of argument that these signals don't interact with subsequent stages in any way. In other words, the output transformer, speaker, and cabinet cannot have any other distortion products interact with these harmonics to produce something similar to aliasing. If that's true, which is not a certainty but we'll assume you've done the research to have some degree of certainty about this, then these supersonic frequencies do not make it past the cabinet, which acts as a lowpass filter anyway and starts rolling off around 5-6kHz.
 
However, in the virtual world, these supersonic signals DO exist and CAN interact with the clock frequency, producing aliasing and audible artifacts not only within the audible range, but within a range that can be reproduced by virtual cabinets that emulate the frequency response of physical cabinets.
 
Oversampling pushes the clock frequency high enough that these supersonic signals become far less relevant, and create few if any audible distortion products. The same phenomenon occurs with running at 96kHz.
 
Again, it's not about high frequencies we hear. It's about high frequencies that cause artifacts. You keep coming back to "well a speaker couldn't reproduce those frequencies" which has nothing to do with what's happening inside the computer.
 

So Craig, who's going to be on that panel with you? Any real engineers?

 
The panelists will be Alan Silverman, Steve Guttenberg, Jerry Harrison, Leo Hoarty, and myself, with Michael Fremen as moderator.
 

bitflipper
Your best candidate for justifying higher sample rates probably lies in virtual instruments and effects that can create frequencies beyond Nyquist: software synthesizers, distortion plugins and fast-attack dynamics processors.

 
I thought it was clear at the very outset that's what this thread was all about, and why I started it. If you've been under the assumption that WASN'T what the thread was about, then your comments make more sense.
 
 

Even in those cases where a processor or an oscillator might generate supersonic content, a mere doubling of the sample rate isn't an effective way to deal with it.

 
An audible improvement is by definition effective. I've played comparison files for many people and they all hear the difference. Jeff Evans ran his own tests with different gear and heard an improvement. I can pick out which file is which 100% of the time. So can Steve Fortner from Keyboard magazine. None of this "well, better than chance." There are plenty of references on the web that justify the math behind what kind of audible artifacts you're likely to obtain at various sample rates, and their intensity; 96kHz is an improvement over 44.1kHz yet remains relatively cost-effective.
 

While there will be fewer frequencies to alias, there'll still be plenty of potentially harmful harmonics to make trouble.

 
But you said "Let's look at a practical example, an electric guitar played through a high-gain amp sim...I didn't do the math, but the level of the 17th harmonic is going to be down more than 90 dB from the fundamental. IOW, inaudible." So which is it? Are the harmonics from distortion going to be "inaudible," or have "plenty of potentially harmful harmonics to make trouble?" 
 
I think the answer lies in between, which is why simply doubling the sample rate produces an audible improvement.
 

You'd really need to at least quadruple the sample rate to be effective, but nobody's making a case for 192 KHz sample rates.

 
Actually, there are people within the AES, CEA, and the record industry debating whether 96k is enough so it is incorrect that "nobody's making a case for it." These involve some very heated discussions by people who have impeccable academic credentials in digital audio engineering. Unfortunately, 192kHz really cuts down on the ability of digital audio interfaces to stream audio using existing computer ports (although I haven't stress-tested Thunderbolt yet). Just as 96kHz was not practical when the CD was invented and so those wanting a higher sample rate than 44.1kHz were overruled, the same could happen with 192kHz, and for the same reasons. Someone who records only classical music could completely justify the argument that 44.1kHz is all that's needed. NIN might have a harder time...so how much should consumers have to pay if they only listen to NIN? Or only to classical music? Tradeoffs must be made in the real world. 
 

I'm not saying that there necessarily isn't some kind of justification for 96 KHz. But over the years people have tried and tried and grasped at increasingly unlikely straws to make the case and so far nobody's come up with a rationale that's supported by science.

 
The premise of this thread is supported by the science of aliasing, Nyquist, clock frequencies, foldover distortion, the harmonics generated by particular processes internal to the computer, and the effects of raising sampling rates through higher clock speeds or oversampling. I think those provide more than enough scientific rationale. There is both practical justification for what I claimed and theoretical justification that will remain in place as long as the concepts behind aliasing, Nyquist, clock frequencies, foldover distortion, the harmonics generated by particular processes internal to the computer, and the effects of raising sampling rates through higher clock speeds or oversampling are generally accepted by mainstream science.
 
There are recording/mastering engineers who make platinum records and hear the difference. There are audio engineers who are grounded in digital audio theory and can confirm what the recording/mastering engineers hear. The final piece of the puzzle is that the lay people I've played comparative files for can reliably detect a difference (as can a pro like Jeff Evans), and agree which recording protocol produces superior high-frequency reproduction with particular types of program material. That kinda nails it shut as far as I'm concerned.
 

Dave have you downloaded the first two files from my post on page #4 of this thread. (please let me know if the links run out and I will generate two new ones to extend the time)
 
I believe the differences here are very obvious and more so than in Craig's original comparison although I can hear it there as well.
 
In my case because 'Prism' is so into producing such rich and detailed high end I see that as a reason why the two files are quite different in their outcome. The 96K rendered down version is just smoother and sweeter and sounds much more like it did at the time.
 
What is all the extra high end that appears in the 44.1K version. Where is it all coming from?
 
Note BTW right at the end of this example notice how I am holding a very smooth lush chord that sounds quite analog in its sound and it does not have too much going on up top. Notice at this point how the two versions of this file are almost the same or very similar. Both have a very smooth sound at that point. I think the reason is due to the fact I just got both 'Prism' and 'Wavestation' producing a very smooth sound at that point with not much going on up high. (probably in my playing) So it really does depend on what the actual VST synth is up to in terms of the type of sound it is making.
 
When I do this for example with my Oberheim VST (Sonic Projects OPX Pro Mk II) just producing a very warm fat brass or string patch the differences become almost inaudible.  Now we are talking analog synth mode with the filter cutoff down low so there is probably nothing above about 6 or 7 kHz!
 
But 'Prism' creating a very top end rich set of harmonics (that are being modulated to move horizontally remember!) in a given patch on the other hand sends the resultant rendered wave into a frenzy at 44.1K. I believe the difference even at 96K is staggering and well worth the increase in quality as a result of going up to 96K. The benefits are there to be heard.
 
I might try it at 192 kHz but I suspect the differences between that and 96K might be very small and certainly not worth all the extra effort required on the computer's behalf just managing that sample rate alone.