Latency control on Transpose-feature

  • Quote

    It doesn;t know to transpose A to Ab if the A hasn't been played yet.

    as i said: it just needs to think: everything that comes in, i will let out through the -100Cents door.

    it does not need to know an exact note. just put every signal 100 cents(for example) down.


    the digitech drop at least has no feelable latency when going down 1 or 2 halfsteps, it gets a tiny little washy when going 3 steps down. i wouldn't use it any lower.

  • as i said: it just needs to think: everything that comes in, i will let out through the -100Cents door.

    it does not need to know an exact note. just put every signal 100 cents(for example) down.


    the digitech drop at least has no feelable latency when going down 1 or 2 halfsteps, it gets a tiny little washy when going 3 steps down. i wouldn't use it any lower.

    I'm not sure how it's done in the Kemper, but you are correct that a direct pitch shift (i.e. minus 1 half step) should have very little latency because it doesn't need to detect the input pitch. However, if you are using an intelligent pitch shifter (i.e. shifting withing the scale of a key) then there will be more latency as the algorithm must detect the pitch before it can apply the appropriate shift to stay within the desired scale.

  • Quote

    I'm not sure how it's done in the Kemper, but you are correct that a direct pitch shift (i.e. minus 1 half step) should have very little latency because it doesn't need to detect the input pitch. However, if you are using an intelligent pitch shifter (i.e. shifting withing the scale of a key) then there will be more latency as the algorithm must detect the pitch before it can apply the appropriate shift to stay within the desired scale

    YES! Thank you. That's what i mean. A fixed pitch doesn't require great calculations as a harmonized pitch

  • as i said: it just needs to think: everything that comes in, i will let out through the -100Cents door.

    it does not need to know an exact note. just put every signal 100 cents(for example) down.


    the digitech drop at least has no feelable latency when going down 1 or 2 halfsteps, it gets a tiny little washy when going 3 steps down. i wouldn't use it any lower.

    Unfortunately, that's not how pitch shift works.


    As far as I know you can't do what you are suggesting with a digital pitch shift. The math is way above my head but hear are a few papers that explain it. One relates to real time pitch transpose on guitar (less than 50ms). Another the different ways of converting pitch and what constitutes low latency (less than 20ms) while the last on talks about midi pitch detection but the sampling of the pitch is the same whether using it for a transpose algorithm or to convert to a midi note value and confirms the MINIMUM latency for a low E on guitar is 12ms plus the time taken to run the algorythm.


    building a pitchshift pedal


    low latency pitch shifting


    very low latency pitch tracker


    pitch shifting is not a simple thing to do.


    I'm not convinced that the latency gets any greater the further you transpose from the original note. I expect it is the changes in formants and digital artifacts that degrade the sound and make us think we are experiencing latency but I could be wrong.

  • You are correct that there will always be a processing latency in any digital pitch shifting algorithm. However, the latency for a chromatic pitch shifter will/should be less than a diatonic (harmonic) pitch shifter as the later must detect the input pitch to apply the correct shift interval. I haven't investigated myself, but I would assume that the transpose option in rig settings will have less latency than the chromatic pitch shifter. I'll have to investigate this evening.

  • I don't want to offend anyone.

    It is probably a sign of today's times that subjects related to mathematics, programming and algorithms are expressed by people who have no clue about it. They imagine how it should work and believe that they are right even without reading even one scientific paper on the subject.

    https://dafx.org/understanding…ronous-overlap-add-psola/

  • I don't want to offend anyone.

    It is probably a sign of today's times that subjects related to mathematics, programming and algorithms are expressed by people who have no clue about it. They imagine how it should work and believe that they are right even without reading even one scientific paper on the subject.

    https://dafx.org/understanding…ronous-overlap-add-psola/

    Yep. Hence there is always latency when applying digital pitch shifting. However, I stand by my assertion that input pitch recognition is not needed for a chromatic pitch shifter. Only an envelope detector is required which requires less samples of the digitized waveform then a pitch detector.


    I speculate that the harmonic pitch shifter, on the other hand, requires both an envelope detector and a frequency detector. Both pitch shifters will then use some kind of algorithm similar to the one you linked.


    Edit... After considering this a little more, I'm definitely minimizing the complexity of the process. For any pitch shifting, some sort of frequency analysis most likely takes place to provide a realistic sounding shift. I've talked myself out of this argument (don't tell my wife!)

  • I have a genius thought that I'd like to share.


    Why don't we actually ask CK to comment on this so he can explain his thoughts and how it pertains to the KPA hardware/processor etc etc...


    Paging @CK, paging @CK - can you please add your thoughts to this thread so we can All understand more about it from someone who knows.


    Thanking you in advance naturally.

  • Edit... After considering this a little more, I'm definitely minimizing the complexity of the process. For any pitch shifting, some sort of frequency analysis most likely takes place to provide a realistic sounding shift. I've talked myself out of this argument (don't tell my wife!)

    yes, that’s the point I was trying to make with the first link I posted which discussed the need to sample and time adjust etc to maintain the original sound rather than come out sounding like a flute or something 😆

  • Surely simply dropping the sample rate by an arropriate percentage for any given chromatic detune (raising it for a rise in pitch) would be all the Kemper would need to do? The data stream could be returned to the internal sampling rate (700kHz) when it leaves the pitch module or stomp slot.


    IOW, no detection should be required for this percentile SR adjustment.

  • We know that the transpose algorithm was improved a year or so ago, so there is probably not a lot more to be gained now. I would certainly like to see a ducking parameter added to transpose, or pitchless transients as Alan describes it. This seems like the only substantial improvement that could be made for the feel in live applications.

  • That would be fantastic, to hear from him ,,but either way.. this has been a great talk, I love that so many folks care enough to add to it..

    Thanks again, I programmed my variax for 4 diff tunings,, I'm a happy guy at the moment,,( but it cost 650 bucks) oh well its only money,,and one more thing to lug around,,

  • Surely simply dropping the sample rate by an arropriate percentage for any given chromatic detune (raising it for a rise in pitch) would be all the Kemper would need to do? The data stream could be returned to the internal sampling rate (700kHz) when it leaves the pitch module or stomp slot.


    IOW, no detection should be required for this percentile SR adjustment.

    probably this would sound a lot like the micky mouse effects you got from early cheap sampling Keyboards in the 80´s.

    would be an nice option for the experimental guitarists :)

  • Not trying to be a "wise guy" but Ford Mustang was first released in 1964 ;)


    Cheers !