Thoughts on bandwidth, sample rates and other miscellaneous spewing...

  • Why is it that audio interfaces are developed to support 192KHZ sample rate if 44.1KHZ is equal? Are they wasting time and selling us snake oil?


    Why do plugins upsample internally to higher sample rates when editing to perform calculations?


    If some are ok with compromise, regardless whether it's little or small compromise, that's fine and, if others want to eek the last possible incremental benefit from using higher sample rates, that should also be fine.

  • But if you tell me I’m “wrong” then you’d better have a better track record and discography before I take you seriously.

    Kemper-forum Etiquette Tips


    Hmm... point #7 in my etiquette-tips thread reads:


    7) Treat everyone with (equal) respect. Lack of talent, experience (or both) and lengthy CVs mean nothing here. Try to judge requests and opinions on their merits and not through a lens that filters so-called big guns from noobs. We're a family, a community with several passions in common at least, and all members deserve the best-possible treatment we're able to deliver.


    I'm not telling you what to do; I intended that thread as a means of suggestion only, FWIMBW.


    Besides, one doesn't have to have had hittoons or engineered successful albums in order to have good ears. Good ears, any measure of success in the industry or length of time involved in it are mutually-exclusive.


    Just sayin'. Enjoying reading everyone's input, BTW, so thank you!

  • This P*%$ing match is starting to remind me of the Groucho Marxs line:


    “Who are you going to believe, me or your own lying eyes?”


    Only insert the word ears for eyes.

    You asked.


    Do the listening test I suggested, see if your ears are lying.


    Why is it that audio interfaces are developed to support 192KHZ sample rate if 44.1KHZ is equal? Are they wasting time and selling us snake oil?


    Why do plugins upsample internally to higher sample rates when editing to perform calculations?


    If some are ok with compromise, regardless whether it's little or small compromise, that's fine and, if others want to eek the last possible incremental benefit from using higher sample rates, that should also be fine.

    Honestly no idea at this point. However there are some different pieces of kit out there that *only* take specific sample rates so your whole pipeline needs to be at that rate. However they tend to just be 48k rather than 44.1k. It may just be "we'd look crap if we had the only convertor that didn't support 192", so they keep it in for that.


    A large part of the 192 thing from what I could see was a race to try and capitalize on peoples misconception that higher sampling frequency meant better, this was also at the time when everyone was getting a big ben word clock for their convertors and swearing blind how much better things sounded, which maybe they did (jitter etc, but not such a big deal anymore), but not any more. It was also happening back when most ITB effects showed really noticeable aliasing, and it was a way of differentiating high end and low end convertors briefly. Since then things have changed, most high end effects now supersample, convertors have improved dramatically, even the analog side has improved no end with very clean op-amps becoming dirt cheap. Maybe someone out there has a real world use case for 192.

  • I too have always had the impression that it was an obvious marketing angle begging to be exploited, something akin to horsepower / kilowatts for car sales.


    Also, I haven't heard it mentioned that every interface model out there that offers multiple rates has a sweet spot, the point at which it converts most-optimally. This IMHO is something I can't get out of my mind every time I read about this dude or that who's recorded at multiple rates through the same interface and compared the results. In these cases, who's to say where the sweet spot lies? In mid to high-end models I suspect it's usually in the 88.2 / 96kHz area, whereas in the cheapest interfaces, which of course are universally ignored in such tests, I dare say it's more likely to be in the 44.1 / 48kHz range.


    This aspect of the picture plays into the hands of those who prefer higher rates as well as those selling interfaces that offer them. "Hey, my [insert esoteric high-end interface name] sounds way better at 96kHz than 44.1". Well, if that's where its sweet spot is, one would expect that, no?


    Just sayin'.

  • Honestly more likely that in the past a low end interface would do better at higher sampling rates and high end at lower. Mostly due to the quality of the bandpass filters - it takes good quality to do "low" frequency reconstruction.


    These days there's so little between them all that even Apogee have got on board the DSP effects bandwagon to help differentiate their offerings from the entry level stuff and justify their price tags.

  • Hold on a minute......... Had a couple drinks for Easter so I'll read the best of this and understand it better tomorrow but looks like a very good topic to enlighten me. Thanks ahead of time! Remember X's and O's in the digital world are just that.

  • POD 2.0 sampling rate was 32KHZ and based on science you can argue that a guitar processor doesn't need anymore than that. right?


    Then why is the internal sampling rate of the Kemper is over 700KHZ?? if higher sampling rates don't add any benefits?


    Was this just another marketing gimmick then? I'm genuinely interested in a reasonable explanation.


    this is from an interview with CK



    Can you talk about what’s under the hood (processors, speeds, sampling rate, A/D conversion etc)?
    The main DSP is a Freescale DSP (formerly Motorola) running at an equivalent of 400 MHz speed. The code consists of tens of thousands of lines of pure assembler code. The global sampling rate is 44.1 kHz, while the internal sampling rate is partially much higher. The algorithm for the tube simulation runs on more than 700 kHz sampling rate (!).



    https://www.guitar-muse.com/kemper-profiling-amp-2949-2949

  • Easy answer, Deano. As I only have a layman's understanding of this, here's my explanation in low-brow, layman's terms:


    At every stage / step in a complex set of calculations, the results that are passed on to the next step are rounded to the nearest smallest figure the chip/s can deal with. IOW, 0.0000006 might become 0.000001, and so on. The more an initial figure is processed the greater the error becomes.


    With "simple" AD/DA conversion, there's a minimal amount of this going on, but in cases of FX algorithms and modelling routines many consecutive calculations can lead to audible errors (distortion of the "pure" waveform). The greater the accuracy, "decimal-point-wise", that a routine can be run with, the smaller the resulting errors will be when the figures are rounded back down to "normal", in this case, sampling rates. This is why plugins and CK's algorithms "upsample" the waveforms in order to be able to perform complex manipulations without leaving audible distortion footprints before conversion back down to regular rates such as those we might use in our DAW's.


    You see this sort of thing when you mess with a basic calculator, where you might end up with a figure of 4.999999999 but you know that in fact the true answer is 5. Punch in 10 divided by 9 and you'll get 1.111111111, for example, but you know that the decimal places go on to infinity 'cause it's a recurring figure. The calculator can't display the value with any more accuracy than that in this case.

  • The reason you need a higher sample rate (supersampling) within your FX is what I've mentioned earlier in this thread - aliasing artifacts.


    In order to accurately reproduce an audio waveform you need to sample it at 2x or above the maximum frequency you want to recreate.


    That means you only need 44.1k on your input and your output likewise understands and reconstructs the waveform from these samples accurately up to and not above half of the frequency.


    This part is pretty easy to understand. Now it gets complex.


    Throughout the processing of this signal you must have a bandpass that eliminates all frequencies above the so called Nyquist frequency from the samples. The trouble with that is of course they're just samples at a set rate, how would it know if there's a frequency in there that's higher? The highest you can possible accurately work out is a waveform with a peak and a trough on every other sample i.e. 1, -1, 1, -1, 1, -1... etc.


    That means that any synth that adds frequencies in to the waveform needs to ensure it's signal is bandpassed to below half of the sample rate. But what about effects? Things like compression? That makes the waveform more squared off, and if you know anything about DSP you know that a truly square waveform is a sum of all frequencies. So it's introducing frequencies in that are higher than nyquist, the result - aliasing artifacts, it can't possibly generate frequencies higher than 2x the sample rate so instead of interpreting those corners as being higher than 2x it creates a waveform with frequencies below nyquist that goes through it, you get harmonics beneath the signal. But there's no possible way of band limiting anything above the frequency because of the sample rate.... so how do you solve this? You supersample - 10x at least, this means you virtually reconstruct the waveform as is. Then you apply your effect, e.g. a compressor. Then you bandpass this, because decimating back to the 44.1k you were using before. This eliminates the audible aliasing artifacts. 10x is considered the minimum to bring the harmonics created by aliasing artifacts out of human hearing range.


    The trouble with the Pod is that it did not supersample. So the 32k was only high enough to reproduce a signal up to 16khz provided there was no distortion etc on it. If the algorithm had supersampled internally then it would be fine. The only limitations then would be the algorithm itself, i.e. a bad sounding distortion isn't going to sound any less "bad" at 700k than at 32k.


    AFAIK - when Christoph is talking about operating at 700k he's talking about the supersampled algorithm internally, not the AD/DA conversion of the Kemper. As the SPDIF was limited to 44.1k for all this time (and were you hearing aliasing artifacts during that time?) it's probably safe to assume that the AD is also set to 44.1k.


    Higher sample rates don't automatically make for a better quality signal within the human hearing range. On older bad quality convertors doubling the rate to improved the relative Q of the bandpass was a common thing, hence they would sound better at higher rates, but I'm pretty sure that's no longer a thing.

  • If Sony tomorrow comes out soon and introduced a player that has ridiculously high sampling rate player in the thousands let's say, and we realize that it's so much better than CD quality, even though CD quality seems to be good enough as many aren't complaining, would that mean that all the music recorded at 44.1KHZ was really lacking fundamentally.


    Is it possible that the whole industry is keeping this 44.1KHZ standard is really crappy if compared with much higher sample rates but the only reason it's kept is because it seems good enough and money is being made Period.


    I do remember reading years ago that Sony was planing such a player but it seems that CD quality a really outdated format is still alive and well and is here to stay for the foreseeable future just because it's barely adequate in comparison to what's actually possible.


    In either case in multi track situations a marginal improvement can be significant and to the summing bus.

  • Yeah, Sony actually did that eons ago, Deano, when it invented DSD (Direct Steam Digital).


    Operates at around 3+MHz IIRC. Some audio software has offered features to take advantage of it I think, but it took over a decade for anyone to jump onboard.


    Essentially, instead of taking x number of 16 / 24 / 32-bit snapshots of audio level per second, it records only the changes encountered since the previous level captured, 1 bit at a time, but very-fast. So, "up a tiny bit... down a tiny bit", as opposed to "this is the actual level... this is the actual level".

  • Why is it that audio interfaces are developed to support 192KHZ sample rate if 44.1KHZ is equal? Are they wasting time and selling us snake oil?


    Why do plugins upsample internally to higher sample rates when editing to perform calculations?


    If some are ok with compromise, regardless whether it's little or small compromise, that's fine and, if others want to eek the last possible incremental benefit from using higher sample rates, that should also be fine.


    Yes, they are selling snake oil !
    But they do it by the request of the users for snake oil.
    The industry provide high sample rate simply because one started and the others had to follow.
    And it‘s not a big challenge to have audio interfaces and DAWs run on high sample rates, compared to what is gained by the marketing effect.
    Please check: No hardware or software maker ever recomments the use of higher sample rate still.


    Beeing in this industry for more than 20 years, I have not met any other industry member that can tell about superior quaility of high sample rates.



    You can tell by this and other discussions on the internet that it‘s impossible to fight the confirmation bias, old paradigms and misconceptions. Thus the majority of the industry stay silent and give their customers what they request and the choice to waste half of their precious computer power.


    By arguing here on the forum I am not really caring about my business, obviously.
    See, I could instead stay silent, go back to work and create something with some real or pseudo high sample rate features, market it and charge €100 more (for snake oil).


  • The Profiler forum might not be the right place to dismiss A/B comparisons
    So you think a valid blind test for different sample rates is not suitable for you or average home studio users that spend a lot of money for equipment?



    I have found a quote from George Massenburg from 20 years ago:


    Empirically, have you found that recordings made at 88.2 kHz or 96 kHz sample rates sound better?
    Yeah, I think so. I hear more high end. But "empirical" means more than just a listening test. What I need to do is make recordings and have them be a part of my life for a time. It's the texture - lace-filigree delicacy of a performance, ambiguities in playback - that, over time, will fill in a sound picture.


    Is there any later cites or more scientific approaches from the master?


    Lavry does not claim deep reasons for using high sample rates. Instead he recomments not to go too high because of increasing nonlinearities. But I could not see him explaining the backrground of his thesis. Do I miss something?


    In my knowledge high sample rates do not harm the audio quality whatsoever.
    It only harms your wallet.

  • Yes, they are selling snake oil !

    That's all I need to know. Thank you very much, I truly appreciate your honesty and I trust you 100% as an expert in this field. The KPA at 44.1 sounds better than anything I heard at much higher sampling rates so the only variable that really matters is who's writing the code and developing the product.

  • Yes, they are selling snake oil !
    But they do it by the request of the users for snake oil.
    The industry provide high sample rate simply because one started and the others had to follow.
    And it‘s not a big challenge to have audio interfaces and DAWs run on high sample rates, compared to what is gained by the marketing effect.
    Please check: No hardware or software maker ever recomments the use of higher sample rate still.

    I agree partly - I think your observations about marketing effect are true.


    However, for SOME applications a higher sample rate is beneficial, for instance if you do audio stretching, say for sound design purposes (and maybe there's a benefit in editing too, like time correcting vocals, quantising drums etc?).

  • I agree partly - I think your observations about marketing effect are true.
    However, for SOME applications a higher sample rate is beneficial, for instance if you do audio stretching, say for sound design purposes (and maybe there's a benefit in editing too, like time correcting vocals, quantising drums etc?).


    Yes, for extreme sound design it can be useful, in the case you create artificial sound by pitching a sample down by more than an octave, and still appreciate the full spectrum, which might rarely be the case.


    Check our pitch shifter and pitch down your guitar by one or two octaves. You will get very interesting results, and it will not be the first thing coming in your mind that you might miss parts of the upper spectrum. At least it was never mentioned on this forum.


    Wen stretching or quantizing audio, you will not shift the spectrum at all.
    When you do voice correction, you will presumably use a plug-in with formant preservation. This will also keep the frequency responce where it was, fortunately.

  • Yes, for extreme sound design it can be useful, in the case you create artificial sound by pitching a sample down by more than an octave, and still appreciate the full spectrum, which might rarely be the case.


    Check our pitch shifter and pitch down your guitar by one or two octaves. You will get very interesting results, and it will not be the first thing coming in your mind that you might miss parts of the upper spectrum. At least it was never mentioned on this forum.


    Wen stretching or quantizing audio, you will not shift the spectrum at all.
    When you do voice correction, you will presumably use a plug-in with formant preservation. This will also keep the frequency responce where it was, fortunately.

    Is it not possible to do real-time formant preservation as Auto-Tune live does? Or some other method of either maintaining or simulating the transient/noise/aharmonic elements of the sound and only shifting the fundamental? Even just working out the noticeable harmonic falloff along frequency and artificially synthesizing an extension of it on top?


    It’s pretty edge use case stuff though.

  • Our pitch shifters with Formant Freeze activated perform real time formant preservation similar to Auto Tune.
    Proud to say it‘s the first and only implementation for the guitar world so far.
    However, it only works with monophonic input, like Auto Tune does.


    Still I don‘t consider it to be an edge use case, as it makes the shifted string sound natural.