Longer IR length support

  • This goes hand in hand with Bommel 's recent request. I'd love to see .kipr support for longer IR lengths. My IRs just seem to sound better instantiated in the DAW. I think this could be attributable to sample length. The A*e FXII supports IRs up to 4096 samples. The He*ix, up to 2048 samples. Some of my acoustic guitar IRs are 8192 samples+.


    Kemper team, if this is not a CPU limitation, I would be so grateful to see longer IR support implemented in OS 6.0! Thanks for all you do.

  • Here follow some notes on the matter from J. Mitchell (TGP).
    Warning: quite long, very interesting (and enlightening) tho.

    HTH


  • I truncated Bass cab IRs and Guitar cab IRs on close mics. I would say bright guitar cabs are cool at even 1024samples when they are close miced. Even small rooms work okayish. Right now i experiment with stereo IRs and Stereo Room micing ... on rooms either a offset or a IR file without minimum phase transformation, so, with the natural predelay are ultra important. I made IRs in a church too and soon i can show you some blind tests about truncated and not truncated files and so on ...


    Overall i think longer IRs and more at once sound fantastic, i hope the kemper guys listen to us, it is simply sooo important

  • It's important to understand that Kemper cabs are transcoded. Even when using CabMaker, the impulse response you give it will not result in the Kemper storing that identical impulse. Christoph has never revealed exactly what the Kemper cab process is, but it does result in fairly short minphased responses with some kind of tweak possibly for the benefit of bass correction or simulated ambience. CabMaker actually allows quite generous impulse responses in terms of length; I can't recall at the moment, possibly 1.5 sec or more - but in the end they're all transcoded Kemper style and at the very least, all time-related information (such as early reflections) is lost.

    I did a bunch of filter analyses with a dedicated speaker-analysis tool back in the early days just to simulate the effect of impulse truncation in the frequency domain. I disagree with the earlier post that 20-40mS is sufficient if accurate bass response is your goal, even for guitar purposes; 100mS (about 4K samples @ 44.1K) would be my preference, but of course most hardware solutions skimp with half that and get away with it.

  • The best way to convert a cab IR of a certain length and sample rate to a cab IR of different length and/or sample rate is definitely not to truncate (and/or resample) the original. If it's just the EQ you are after -- i.e. time-based information is of no importance as with pure cabs (no predelay, no reflections) -- you better construct the filter taps of the new IR from the frequency response of the original.


    And regarding the filter length: Anything audible beyond a few 10ms will be perceived as being echoed/reverberated/smeared. Not to mention that large filters consume more computing power. So it is best to have your IR length as short as possible (i.e. with reasonable deviations from the desired frequency response).


    If the frequency response curve of the original IR is sufficiently smooth 2048 filter taps @44.1kHz will accurately sample the curve at 10.77Hz intervals. I don't know if the Kemper Team follows the approach discussed here but I bet I may create [email protected] IRs from your original bass cab IRs which are considerably better than the truncated versions. :S


    Cheers,

    René

  • Sorry for reactivating this thread, but I've enhanced my tools for analysing and recreating IRs quit a bit (see attached plots for a hint) and would appreciate them to be really challenged. I've collected quite a few bass cab IRs but never found anything like the IR presented here.


    So Bommel, may I carefully ask whether it is possible to get the exceptionally long bass cab IR you've presented above? Just post it here or PM me if you like. In exchange I would send back recreated short versions (wether good or not) and present the analysis results here.


    Cheers-

  • Good Bommel was kind enough to help me analyse the bass cab IR he presented above (Thank you so much!). I've recreated shorter versions of said IR from the frequency response of the original and compared the results to just truncated versions of equal lengths. Attached you can find comparison plots for the frequency response and the deviation from the original for IR lengths of 4096, 2048, 1536, 1024 and 768 samples.


    I leave the review of the recreated IRs to Bommel himself. But I wouldn't be surprised if the differences between even the recreated 1536 sample version and the original can only be heard in direct A/B comparisons.


    In conclusion I'd say that if an IR is reluctant to truncation it's worth recreating shorter versions from the frequency response.


    Hope you find this analysis informative:) or even helpful:S. Cheers-

  • The reason for this particular IR to fight so hard against truncation is the steep cut-off at 33Hz. Such a curve can only be accurately reproduced by filters of extreme length.


    So to drive things even further I tweaked my machinery to allow for greater deviations in the range from 1 to 32Hz. This helps the algorithm to be more precise for frequencies above 32Hz. Attached you can find the results for an IR length of 1024 samples. If frequencies below 33Hz are of no importance or get cut by a filter anyway, you may get away with such an extremely shortened IR.


    Please tell us what you think. Cheers-