Advanced sampling advice: short loops

I am rediscovering the long lost art of creative sampling and perhaps some of you folks can help me out here.

I am trying to figure out what is the best way to make really short (single or perhaps 10 cycle) loopable waveforms. My loops are now usually 1 second or so long and by visually (and by ear) looping them I can get descents enough sustainable loops for my samples.

What I am trying to do now is to synthesise / sample extremely short samples: single or perhaps a dozen or so cycles which I want to use to create new programs. At 48 Khz sampling rate a A#3 would be about 103 samples.

Here's my problem: the sample length of a single cycle never matches with a round number of samples. The A#3 at 48 khz would theoretically have to be 102.968 samples. Which is not possible of course. How do I get around this?

I can think of two ways:

* Sample 10 or so cycles - so you won't notice the minute deviation.
* Use single cycles and detune them a little.

The first method seems the easiest way of solving it my problem; even though the sample would still not be perfect.

The second method seems a little more difficult: I would need to sample a slightly detuned note, loop it and set the 'tune' within the sample.

Anybody have any good tips for this?

Also, how do I figure out how much to 'offset' the tune parameter in the MPC.

For example if I make a single cycle oscillator for C3 (should be about 261.63 Hz) The 'ideal' length (in samples) of the single cycle loop at 48 Khz sampling rate should be 183.47 samples which is impossible. So I create a single cycle loop of 183 samples (or 184) which is a little 'off'.

Playing a 183 sample loop at 48khz would give me a 262,30 Herz tone which is slightly off (0,26% in Hrz) how do I not set the tune parameter?

How can I calculate waht I need to set the 'tune' parameter to in the MPC4000: it's not -00.26 I believe since the tune scales 1 semi note to 100% doesn't it? A semi note is roughly 5,61% off (in its frequency) from the original note. So 0.26%....would that be 0.26/5.61*100 = .046

So tune = -0.05?

Actually that just doesn't sound right. Assuming that the 'tune' parameter is logarithmic too I've come up with the following table:

http://bayimg.com/CaHBpAaBp

Would this be correct?

(I know that especially with lower notes the differences can be very small, but single cycle oscillators at higher frequencies can be off quite a bit)

wow,

I can't help you but it'd be nice to know these things, and to know exactly what you're talking about!! lol

Well, I am trying to figure out how to correct the tuning of a single cycle wave form. (those are extremely short looped samples; the shortest loops possible).

Since by the nature of sampling these loops are always a little out of tune you need to tune them using the tune function of the MPC. And I am trying to figure out how to calculate by how much.

I think I've figured it out (almost).

Can someone verify my maths?

yes, I understand what you are sampling and what the single cycle wavs are, but when you start talking about tuning and such I get lost, but this is something I am trying to learn more about so I am watching this thread.

saintjoe wrote:yes, I understand what you are sampling and what the single cycle wavs are, but when you start talking about tuning and such I get lost, but this is something I am trying to learn more about so I am watching this thread.

Well, the length of the the cycle determines the fundamental pitch (when it's played at its default frequency of 48 khz). The length of a sample never is exactly right (it's always a little off) this will detune the sample a little (upward or downwards) That's why you have to 'adjust' a little by using the tune function.

I am trying to figure out how you can calculate the best length of the sample and the amount of relative tuning that you need to do (that's in the table)...

I wish someone could look at the maths of this for me.

I've done some experimentation last night...using Soundforge...I notice that looping samples in Aksys must be done by right clicking on them and setting the sample itself to loop (in the little window where you also select it's key). If you do it in the big window (where you see the 4 layers of the sample) and click 'loop' there the loop will 'tick'.

hmmmm, good advice, does it loop properly or do you still need to set the loop points?

saintjoe wrote:hmmmm, good advice, does it loop properly or do you still need to set the loop points?

If you make the loop correctly in Soundforge it will loop seemlessly like this. I've also figured out a working method (I think) for creating single cycle loops...I've looked a bit through the documentation of the Arch_Uni_Demo files.

http://www.mpc-forums.com/viewtopic.php?p=770722#770722

These files are all exactly 2048 samples. Why is this good? You can resample them in soundforge to be exactly 1024, 512, 256, 128 etc. samples long. If played at 44.1 KHz these will be almost F (you need to detune them by +23 cents) and if played at 48 KHz they will almost be the F# (you need to detune them -23 cents) in their respective octaves.

So basically ALL you need is to have a (*any*) soundloop that is 2048 samples long and you're good to go.

When I am done experimenting I will write a short tutorial too...

thanks!!!, yeah I was looking at those same demo files, I plan to get the whole thing, but thank you for explaining some of this, it is really helpful to some things I want to get into, basically sound design from this level of creation, so i'm trying to learn as much as I can and you're helping me out

now I just need to purchase a copy of soundforge lol or wavlab.....!!!!

Came in a little late, but the 2048 sample method is good for single cycle waveforms. That Architecture Waveforms pack that was mentioned recently has all of its waveforms the same way. One thing to keep in mind is that the Z engine will only transpose 3 octaves (and 2 in 96k mode). So if you want a bigger range, it makes sense to make a few more samples to span over the keyboard. The higher pitch samples are obviously harder to get perfect since they are shorter, but you can use several cycles to get a longer sample, or just fine tune it. For example, to go up 2 octaves, you could just duplicate it to 4 cycles and then pitch down by 1/4 and still have your 2048 samples.

I def want to play around with this, so far I've only sampled my waveform cycles from synths. This is a lot easier since the cycle is already in whatever pitch you want.

McSmooth wrote:Came in a little late, but the 2048 sample method is good for single cycle waveforms. That Architecture Waveforms pack that was mentioned recently has all of its waveforms the same way. One thing to keep in mind is that the Z engine will only transpose 3 octaves (and 2 in 96k mode). So if you want a bigger range, it makes sense to make a few more samples to span over the keyboard. The higher pitch samples are obviously harder to get perfect since they are shorter, but you can use several cycles to get a longer sample, or just fine tune it. For example, to go up 2 octaves, you could just duplicate it to 4 cycles and then pitch down by 1/4 and still have your 2048 samples.

I def want to play around with this, so far I've only sampled my waveform cycles from synths. This is a lot easier since the cycle is already in whatever pitch you want.

so this wouldn't apply to cycles that you sample from synths? only natural sounds? I plan to get all of my cycles from my synths as well.....I don't have any "real" instruments.

saintjoe wrote:thanks!!!, yeah I was looking at those same demo files, I plan to get the whole thing, but thank you for explaining some of this, it is really helpful to some things I want to get into, basically sound design from this level of creation, so i'm trying to learn as much as I can and you're helping me out

now I just need to purchase a copy of soundforge lol or wavlab.....!!!!

I think it's easier to resample the wavs to half the frequency and then manual set back the playback to 48 (or 44.1 Khz) so you get samples that are just half as short: that's why the 2048 method is so nice: you don't get any aliasing artefacts.

saintjoe wrote:so this wouldn't apply to cycles that you sample from synths? only natural sounds? I plan to get all of my cycles from my synths as well.....I don't have any "real" instruments.

That's the tough one (look at my table) it's really hard to get an exact cycle from a synth: you're always a fraction of a sample 'off'. Unless you can 'tune' the synth by 23 cents and play the F (or the F#, which depends on if you're using 44.1 or 48 khz) from each octave.

In that case: you're good to go.

The trick is to have the cycle length be exactly a rounded number of samples (and preferably one that's a multiple of 2^x)

cool, thanks