Clawfinger said that good IRs sound great at 96khz and more....

mba

Power User
IRs sound great at 96khz and more....

I just tested a Mytek and i agree.
Good IRs give their best at 96khz or more. With lower rates it sounds digital.
I don't know why.

By the way i have to correct what Clawfinger said but it doesn't change my own questions.(i can't change the title of the thread)

Clawfinger said this "we've done some extensive testing with DSP's and it felt bad when we tested any frequency under 96KHz."
All apologize Jocke. ;o)
 
Last edited:
Could you clarify what you mean?

The internal sample rate of the convolution engine? The input and output sampling rate of the device? A DAW plugin?

Richard
 
I just tested a Mytek and i agree.
Good IRs give their best at 96khz or more. With lower rates it sounds digital.
I don't know why.

I'm not sure what you mean with 'sounds digital', and frankly speaking, I don't really believe it. The cab IRs in the Axe sound fine to me. Also, I often use IRs for reverb in DAWs which are 44.1khz and I never noticed any digital artifacts with them.
 
A typical guitar speaker has almost no energy above 10 kHz (even 5 kHz really). So, in fact, 48 kHz is overkill.
 
See this thread. The posts on page include demonstrations.

As far as (24/)96k goes..
You are dealing with several factors that could lead one to (mistakenly) believe this to be true, or to create situations wherein they might appear true. As Cliff has stated, the (practical, applicable) range of reproduction for a (typical) guitar speaker is (far) below 20k; as such, 48k is capable of reproducing any recording accurately. Well, unless you have information that disproves Nyquist/Shannon? If anything, I would look to bit *depth* during the process as a point of debate; greater than the 48k sample rate utilized by the Axe would be.. not even the last place, as there is no question IMO.

If you want to broaden the discussion to include all impulse response/(de)convolution processes/applications, one might open the floor for (constructive) discussion. If you want to discuss the hosting and ADC/DAC involved in hosting and the capture processes, then that is a different, and well discussed, debate. Since you mention "Mytek," could you elaborate on the usage/method(s) utilized in your comparison please? Comparing produced clips @(24/)96k vs (24/)48k could very well be providing a false conclusion that the IR being utilized benefitted from a higher sample rate.
 
  • Like
Reactions: jon
only guys who dont think about the physical/technical facts behind it will state this ..

human ear can hear up to 20k .. so why capture more if we cant here it anyway (48khz is enough according to nyquist and shannon)? there is no magic behind it ... low cutting an 96khz sampled signal to 48khz will sound exactly the same as a directly 48khz recorded signal .. (without looking at errors produces by the low cut)

so listen to nyquist and shannon and get your facts straight ;)
 
Guitar speakers may not. How about other types of IR's, such as acoustic guitars for instance?
Produced frequency content and the ability of an *inherently restricted, frequency based* method of capture and reproduction is a long debated topic. If you believe that frequency content produced by an acoustic guitar "environment" that exceeds ~20k is vital for a recording to capture "properly," then IRs fall within a much, much larger arena of concern (IMO). IOW, no digital recording (*inherently restricted*) will ever accurately capture a performance; each attempt will involve compromise. If one feels that >~20k is negligible/of little/no concern, then, once again, Nyquist/Shannon provide that any properly implemented digital capture and reproduction will be accurate.
 
only guys who dont think about the physical/technical facts behind it will state this ..

human ear can hear up to 20k .. so why capture more if we cant here it anyway (48khz is enough according to nyquist and shannon)? there is no magic behind it ... low cutting an 96khz sampled signal to 48khz will sound exactly the same as a directly 48khz recorded signal .. (without looking at errors produces by the low cut)

so listen to nyquist and shannon and get your facts straight ;)
Exactly. :D
Sort of.. lol..

To elaborate, while Nyquist/Shannon provides the foundation, implementation in "our world" has thus far been "imperfect." If we consider the top end of our hearing to top around 20k, and opt to err in favor of "super hearers," then we could opt to go for even ~22k. The issue faced is in filtering (primarily, superficially, but not solely). To provide some "cushion," opting for a 48k rate allows for certain compromises, including our "imperfects." Internally, however, the math, simply, "works." But, we also face other considerations that do not solely, or necessarily primarily, depend upon our "range of hearing."

For the most part, your statement is 100% true. Unless one dives into that rabbit hole, the blanket statement applies. A good first step, IMO, is Nika Aldrich's book, Digital Audio Explained". IMO, it is the friendliest, "layman's terms," entry level book concerning all things digital audio. It is not a "Saturday read" though.

For a glimpse of the edges of this Rabbit Hole, I think this thread is a fun read :D
 
This actually sounds like it could be an interesting thread. I have a relatively fair grasp of Nyquist and general knowledge about digital audio, but I am far from an expert. I grasp general concepts, the math pretty much loses me when things go way beyond that.

Anyway the thing that bugs the living crap out of me about this whole thing in terms of sample rate is that I have heard on more than one occasion the same exact signal recorded at 44.1kHz and at 96kHz with the same gear and can hear the difference. I'm not talking about "I've got golden ears and I'm special" or any of that crap. I can hear it and if anyone listened to it they would as well. I'm talking night and day differences. And this is with the same bit depth (which in my experience is more critical to quality).

Now the latest experience I had was with that poor 11R I had. I would record at 96kHz because everything else sounded like mud coming out of it. This was using SPDIF out of the 11R. The only thing that made any sense to me was that possibly it used 96kHz natively internally and that any other sample rate selected was going through an onboard SRC inside the 11R which would explain the crappy quality. But it still bugs me because everything that I understand about digital audio screams that I'm smoking magic mushrooms for hearing a difference.

Conversely I can take the analog output of the AxeFXII and send it into my audio interface and record at 48kHz and at 96kHz and cannot hear a difference at all. I just did this two seconds ago to verify it and it kind of restored my faith in good old Sven.

So anyway, I'm just wondering if anyone has any ideas on this.
 
@Sasha:
There are several possible contributors to what you experienced. If you are playing back a single track with no plugins/external processing, then the converters may be playing the sole part of providing this described difference. Without getting too "complex," the quality of the converters (DAC, in this case) would be the culprit. The better the quality of your DAC,.. with quality being a scale approaching a "theoretic perfect"... the less likely it is that, under stringent double-blind conditions, you will hear any difference.

If you are afforded a "perfect" DAC (for argument's sake), and decide to try different sample rates for a session, and employ various plugin (processes), along with level variations, panning, and summing, then you have now involved a pretty nice set of variables. While certain processes can benefit from a higher sample rate, others do not. IMO, this is an excellent example of "know your tools!" It can/is also the difference between an actual "audio engineer" and a "mixer."

Your experience with your 11R is, IMO, an excellent example of ADC/DAC "imperfection." Higher sample rates in these cases (generally speaking) allow for this rather simple statement: it is less expensive/easier to construct a quality ADC/DAC to operate at higher sample rates. The closer we approach 44.1k (as an accepted target for practically applied Nyquist/Shannon with regard to human hearing range), the more "expensive" the implementation becomes. There is a reason why two channels of Weiss, for example, run over $9k for ADC and DAC... EACH (>$9k two chan ADC, >$9k two chan DAC). The Axe has far superior ADC/DAC than the 11R (please correct me if I am wrong here).

To spend just a little time in that aforementioned Rabbit Hole is time well spent IMO. Shy of that, I would recommend listening to your system. If you feel that (24/)96k sounds better, then do it. Computer hardware is competent and plentiful enough so that compromising in that area rather than forking out mega bucks to grab top quality 48K ADC/DAC is an excellent compromise IMO. However, take some time and compare your plugins and such at each rate as well. It may be a bit more difficult to discern differences with them versus the impact of the converters... especially if you are still using those converters. Personally, I work at 24/48k, sometimes 32-f/48k, but for final captures/primary tracking, I prefer 96k. I would not opt for a higher rate.

Sidebar: optimally, I agree with Dan Lavry's general statements, and find reason to desire a ~64k rate. I think it is an excellent compromise, given all the known variables, components, etc we have today. And, apologies for my.. "enthusiasm." This is a "pet area" of mine, and I can get going.. lol...
 
I thought that the whole point of higher sample rates and bit depths was less about what the human ear can decipher and more about minimising the introduction of artefacts when the audio is being processed [because the resolution is so high]..
from what I understand, the human ears can't really tell the difference from 12 bit / 40KHz and up...
however... when you start goofing with the audio [eq, fx etc] you're throwing a pile of math at the binary.. and this is where you'll start hearing nasties like quantisation errors and artefacts in the form of spurious clicks and pops..
so the solution is to do all this math at a far higher resolution...

can you really hear the difference between 24/96 and 24/44.1??
I absolutely know I can't..
and my lil' KRK V6 monitors shelf at 20K so I doubt they can hear the difference either...
just to be sure... I'll ask 'em when they're not busy...
 
Last edited:
Im a noob, but 96khz what?

this refers to the sample rate of the digital audio.. 96,000 samples per second..
the sample rate is basically how many times in a second the level of the audio signal was 'measured' so that it can have a numeric value assigned to it..
so you're nice analogue guitar noises are being measured 96,000 times per second and turned into a stream of numbers..
when you add an FX block and twiddle a dial, you're essentially throwing a pile of math at your lil' stream of numbers...

bit depth - 16 bit, 24 bit etc, determines how many different levels [essentially volumes] from minimum [no volume] to maximum [flat out] can be measured..
more bits = more individual volumes..

you have to be aware that the noise coming out of your guitar [the wave form] is a continuously changing 'wave' like ripples on water...
where the biggest 'wave' [your fundamental] is the note you're playing.. but on top of this wave are smaller waves.. and on those are smaller waves again..
it's the nature of these smaller and smaller waves that make your guitar sound like a guitar and not like a trumpet..
the higher sample rate allows more accurate measuring the shape of these smaller and smaller waves...
the large bit depth allows more accurate measuring of the height of these smaller waves...

and at the other end... after all your knob twiddling... these numbers get turned back into electricity, stuffed into a power amp so that the speakers move the air [sound]..
 
Last edited:
Thanks for the info Nikki-k. I already had an idea about some of what you were talking about in your post. From what I've kind of been led to understand from my time spent trying to learn that a lot of the higher sample rate stuff was kind of along the lines of the rage of over-sampling back in the early 90's. I know that you remember. Everyone had 2X and then 4X and eventually it got up to about 64X. I mean that's great and everything, but from what I can figure out the only thing that really did for anyone was allow them to use cheaper parts to create a steeper cutoff for the low pass filters. And to be honest I can see where the advantage would be in the recording side; I don't get it for playback or D>A conversion, but that's besides the point here. I'm just saying that I do get the point that the converters are what make or break a device and that by increasing the sample rate improves quality, but its usually in the name of saving money on the discrete components that add cost first and foremost.

So that does explain (at least for me) the phenomenon with the 11R. Sure I could use other sample rates, but they all sounded like crap (and I'm not just bashing the thing here because it's not that bad) unless you used 96kHz. I mean the options are nice to have, but if it sounds bad then why even have it in there? I suppose that kind of explains why Cliff chose one sample rate to rule them all so to speak.

And yeah, I've done a lot of testing of my gear and my current audio interface does a really good job at various sample rates and like I said earlier, I can't hear the difference between 96kHz and 48kHz. Its the same noise floor, high end, texture, levels, frequency response, whatever. Its the same. But taking a digital signal straight from the 11R at a down-sampled rate sounded horrendously muddy and muffled.


Im a noob, but 96khz what?
And just to back up what clarky was saying about bit-depth and stuff for nellings; I do feel really comfortable with this area. I look at it like someone giving me a ruler to take measurement (which is all you're doing when you digitize something). The sample is like a snapshot of the signal's voltage at a specific point in time and you have to assign a value to it by measuring it. Now if you have something that you have to measure that is between 0-12 inches and use a 12" ruler marked with only inches (pretend that's 8-bit) you can get close. Write down that measurement and then take the next sample, measure it, write it down, etc. Now because you know how many times per second you wrote those measurements down (sample rate) you can recreate them by 'playing' them back at that same rate or speed at which you wrote them down.

Now any measurement your took that was exactly in one inch increment is fine, but once you have stuff in fractions its been averaged and you lose accuracy. Most of the time that's not a huge deal to be honest. Where it really rears its ugly head is when you are taking consecutive measurements where there's no dynamic range (almost no variations in the signal's voltage). Once you start getting similar measurements in a row it becomes this big old nasty flat line which sounds like clipping, but it's usually at very lower volumes. A higher bit depth like 16-bit will give you the equivalent to a ruler with 1/64th increments. Now all those measurements below 1 inch can actually be measured and given a real value other than just 0 or 1 inch so you don't get that straight line.

This is why you may see it said in places that 16-bit gives you -96dB of SNR. Basically that's saying that it can register a value -96dB from 0dBFS (the highest value you can record without clipping). That's like trying to measure 13 inches on our ruler; we can only go up to 12.

In case you haven't noticed this is getting a little long winded. I'll try to get to the point.

8-bit is 256 values because you've got 8 registers in your byte. It can range from 00000000-111111111. The highest value is all ones and anything higher than that is still going to be all ones. That's your classic clipping from overdriving your A>D converter. The other disadvantage with 8-bit is that it can only get down to -48dB for SNR. That sounds pretty decent on paper and if you record something that is constantly loud its not going to be a problem. But if you have quiet passages it will distort. You'll get a bunch of zeros in multiple consecutive bytes (your actual measurements) and it sounds like really quiet digital clipping. I believe that it's called granulation noise. You also have 256 values so there's a lot of averaging going on for every sample. 8-bit sounds like crap, but because it's a smaller byte it does have its applications. Just not in music.

16 bit gives you 65,536 levels, -96dB and that's actually pretty good for most things. Its what CD's use.

But 24-bit gives you over a million levels and -144dB SNR. That's like taking a micrometer to make those measurements. Granulation noise is damn near non-existent. Honestly I've never really needed anything higher than 16-bit until I started getting into re-amping. The signal level is like -40dB at its peaks. That extra resolution at those lower voltage levels makes a world of difference in that case.

Hopefully nothing in there is wrong; like I said I'm more of a concept guy.

Just to try to summarize my opinions; sample rates are cool; look up Nyquist theorem if you're really interested. The jist of it is that your sample rate has to be at least twice the frequency of the highest sound you want to record. Even if someone claims they can hear above 20kHz no musical information is in that range so they can kiss my butt. :) Sample rate will determine the highest frequency you can replicate digitally and as long as you've got good quality A>D converters you can get the full range of human hearing (and more importantly any type of music) at 44.1kHz with no problem at all. And not to get too far off track here, but how many people actually like to hear stuff above about 15kHz anyway? That's like the scan rate of an old TV set. If you can't hear an old TV whining with no volume then you can't hear anything above what a 32kHz sample rate would give you. If you like to listen to an old TV whine then you need professional help.

But to get back to my point; too much emphasis has been placed on sample rate; its all about the bit depth in my book.

Sorry, I was all over the place. :)
 
Last edited:
I thought that the whole point of higher sample rates and bit depths was less about what the human ear can decipher and more about minimising the introduction of artefacts when the audio is being processed [because the resolution is so high]..

I like this question.. a lot :D

This page of the "mega sample rate thread" I linked previously contains a good bit of information/debate concerning this. IMO, one of the most pertinent points proffered within is within this post by Nika. Not to sound like a Fractalyte or "Axe-alterer," (hehehe ;) ), this is one area that I am betting that Cliff has done his homework.. serious emphasis on the HOME part, as in "beyond stuff done at work"... and his implementations take all of these things (and far, far more) into account, and that he has implemented *methods* that reside in the "Oxford" area that Nika cites.

IOW, within the environment that Cliff is working, and with his level of knowledge and practical experience (plus a great dose of "experimental scientist" :D ), the Axe should be entirely capable of achieving professional+ level audio processing. Sidebar: Take a moment to consider what Cliff has done; by creating a solid foundation with so many knowns (constants), the number of variables to consider for implementations are significantly reduced. This is a VERY good thing (Apple, for one, does this, albeit on a broader scale and scope... not the best example, lol), and, IMO, it allows Cliff to place greater focus elsewhere.

Also, some converters (ADC) will perform the initial conversion at a (much) higher sample rate, and then downsample to the (user selected/preset) rate prior to passing the data along (correct me if I am wrong here please, unable to locate proper citation).
 
  • Like
Reactions: jon
Niiki - yeah I'm familiar with this sort of thing.. so when processing your audio, the 'errors' introduced are way out there at the tiniest decimal places so when you drop back to more conventional sample rates it's as if the errors were not there at all..
 
Back
Top Bottom