The Art of Looping (Part 3)
The final part of this series on creating good sample loops. This month it's yesterday's synthesisers that submit their sounds to the digital manipulation of Chris Meyer.
If you want to give your old synths a new lease of life sampling could be the key. Here are some guidelines on applying the theory to the real world.
IN THE FIRST two parts of this series I covered the basics of looping sampled sounds. Now it's time to discuss putting those practices to some constructive use. Rather than provide you with hard and fast rules about how to do things though, I'll try to explain what works for me, after two years of practice, and hope that it helps you establish your own methods.
I usually do things in batches - set up all the memory, do all the samples, trim them all and so on. Aside from it being a strange kink in my personality, this approach gives me the opportunity to keep comparing different samples of a sound side by side, and ensures that I don't miss a step in the process on any one of them. So I'll be covering all the various topics associated with sampling in the order I'd actually do them.
BEFORE WE DELVE any further into the actual sampling process, it's as good a time as any to talk about sample sources. One angle is to try to recreate traditional instruments, such as cellos, pianos, and horns. This usually requires getting hold of the instrument in question (and preferably someone who can play it), a microphone and a good recorder. University music departments tend to be good sources of the former, if you can gain access to them, but the latter are items to spend your time and money on. The process guarantees unique samples, and is ultimately more satisfying than using sample CDs or tapes. If you do want to take the easy way out there are more CDs of sounds intended for sampling appearing on the market all the time.
For those intending to record acoustic instruments for the first time, try recording your own voice sustaining a vowel sound - it's as good a place as any to start making mistakes, and sampled vocals and choirs always seem to be in vogue.
Another source of samples (particularly percussive ones) is anything lying around the house that sounds promising when struck, bowed, caressed, or whatever. I personally think anything dropped an octave (slowed down to half speed) or farther sounds great. There's a greater margin for "error" here too because it's unlikely anyone will accuse you of having a lousy biscuit tin sample - uniqueness is on your side. No doubt you're terminally bored with the phrase "Sample that!" by now. Well, I've got news for you; since I started doing sound effects work, that phrase has become an important part of my philosophy.
The sample source I'm most fond of is other synths. Part of the large turnover in synths is due to the eternal quest for new sounds, but there's another side to it. As it's financially impractical to have one of every synth around just for the handful (or less) of sounds it really does well, it makes a lot of sense to sample them and sell off the original instrument. Throw in the lack of programmability and polyphony and the drifting oscillators of early analogue synths and it becomes a more attractive proposition still. Of course you'll never have the full flexibility of the original, but it still makes a lot of sense.
And synths are great to sample, because they're easy to play and record, and are generally easily borrowed from music shops or friends with a passion for collecting gear. You could even invest a few pounds in hiring anything you can't borrow.
BEFORE MAKING A sample, you have to decide if there's a way to play it that should be used or avoided. A lot of this has to do with what facilities you have for changing the sound after making the sample. Is there a handle rattling on that saucepan you're about to hit? If it's in a frequency range anywhere near where the rest of the sound is, wrap it up or tape it down - no filter in the world is going to get rid of it later. The same goes for the hisses and hums of a synth - try to cure it now. Any other equalisation at this point is fair game too, but while it may be tempting to equalise everything to death before sampling, I prefer to be more cautious and get the straight sound down first. The third option is to take at least two samples - one dry, and one "dosed" with EQ.
Does your sampler have an adjustable filter or one that can be altered with an envelope? If so, sample an overstruck version of the sound (hit it a bit harder, or open the filters a bit more on the synth). You can always calm it down later, but you can rarely beef it back up.
What type of vibrato, tremolo, or other washes to the sound exist? Remember that they will become a permanent part of the sample, and transpose along with the pitch of the sample. Things such as a Leslie or the even wash of a phase shifter on a string synth don't transpose well; natural, random, and/or complex modulation (such as oscillators beating slightly) do. What sort of memory and looping facilities do you have? Sounds other than pure, vibrato-less tones are hard to loop without crossfade looping, and the more complex the modulation, the longer the loop required. You may be best advised to get rid of it first.
"There's a greater margin for 'error' with unusual samples - it's unlikely anyone will accuse you of having a lousy biscuit tin sample."
This same rule applies whether you're thinking about sampling a solo instrument or a whole ensemble - the more instruments playing at once, the more complex the beating, and the harder it will be to loop. Some samplers even have chorus, layer, or detune facilities to thicken sounds up later. If you have crossfade looping and/or lots of memory, however, sample the best sound you've got now - it'll only come out better later, and will be far more complex and natural than you could recreate with your sampler's own facilities.
Two special notes to those sampling other synths, and whose samplers have synthesiser-like envelopes and the like: 1) remember that you can recreate amplifier and filter envelopes with your sampler, and 2) be wary of trying to recreate many of your amplifier and filter envelopes with your sampler. Well, this one deserves a bit more attention. Keep in mind that transposing your sample all over the place will also change how long it takes to play back. Consequently, your envelope rates will be sped up and slowed down along with the sample's pitch. Dropping a sample an octave cuts the speed of all the envelopes in half; raising it an octave doubles them. If they are simply amplifier envelopes or gross filter envelopes (like an exaggerated wah or sweep), keep them wide open when you sample and use your sampler's facilities to recreate them.
On the other hand, keep in mind that your sampler's envelopes and filters may not sound like the synth's own, or may not do things that the synthesiser's would. You don't hear people rhapsodise about the Minimoog's oscillators; they rhapsodise about its unique filter. Many older synths have dedicated envelope-generating hardware that produces exponential curves which sound a lot more natural than the linear ones on the majority of current samplers. Some older synths have unusual features like inverted envelopes that just don't appear on most samplers. In this case, it's better to use the synthesiser's envelopes and filter(s) when sampling, and on playback, and make the sampler's as neutral as possible.
Before creating too many raw samples, try the following experiment: program a sound on your synth that has a fairly interesting and complex oscillator setup and a slow, full filter sweep up and down with a touch of resonance. Sample that and play it without using an envelope on the sampler itself. Next, sample an equal length of the raw oscillator sound with the filter and resonance off. Then try to recreate the original sound with the sampler's envelopes and filter. This will give you an idea of how close you'll be able to get. Chances are you'll find that the envelopes are fine but the resonance doesn't sound right. In any case, you'll find out what you can and what you can't get away with.
If you can get by with using the sampler's own analogue (or digital) processing, you'll have a little more work to do, but you'll use less memory and will end up with samples that are more consistent across the keyboard. I'm personally very particular about a synth's sound and am usually not offended by transposing envelopes in order to capture a sound as accurately as possible.
Finally, remember that you never hear the natural release of the sound from your sampler. Most samples tend to remain in a loop and use their own envelopes to fade them out. Using a sustain-only loop (see The Art of Looping Part 1) with the sound's natural release occurring afterwards is an alternative, though this eats up more memory, but unless you release a key precisely at the end of the loop, the remainder of the loop will play before the note releases.
FAIRLY BASIC SAMPLING topics these, but here are a few tips to help you on your way.
"Envelope rates change with a sample's pitch; dropping a sample an octave cuts the speed of the envelopes in half raising it an octave doubles them."
The higher the sample rate, the greater the bandwidth - that is, the greater the range of frequencies in the signal after sampling. The first cost of narrow-bandwidth sampling is the upper frequency content of the sample. This economy is normally made in the design of the sampler to keep costs down, or in the bandwidth/sampling time wars in the interests of getting the required length of sample albeit at the cost of quality. Most samplers have a nominal sampling rate of around 30kHz, which translates roughly to a bandwidth of 12kHz - bright enough for old subtractive synthesisers; perhaps not so for some of the newer digital demons. Because higher sampling rates do eat up memory, I quite often sample all but the highest notes around 30kHz, and occasionally the highest one at 40kHz or so - it saves a bit of memory for looping. Simple, no?
But now it's "exception to every rule" time, again: if you transpose a sound downwards, you're transposing its bandwidth down, too. Playing a sound sampled at 30kHz down an octave drops the bandwidth to a dull 6kHz. Low note samples are the ones that are most commonly stretched the furthest down, so some consideration must be paid to sampling these around 40kHz instead.
The real issues are how much memory you have, how much of the keyboard range you plan to cover, and how long the sound is. Very few natural instruments or synthesiser patches sound good over five octaves, but there is a constant impulse to fill the whole keyboard regardless. If you know you're only going to use a sound over the two bass octaves (or whatever), you can balance the decreased number of samples off against longer ones. In most cases, though, the sampler's transpose range won't gracefully cover those two octaves and more than one sample will have to be taken. Tuck it into the back of your mind right now that samples tend to transpose down better than they transpose up (the old chipmunk effect), and remember this when it comes time to pick which pitches to sample.
As I mostly play in a studio environment I don't mind having only one or two different sounds on a disk, so I listen to how long a sound takes to evolve from its attack to a steady state (for looping), and decide then if I'm going to try to fit it into half of the memory or if it's going to require all of it. Also, the evolution of the sound and its loop is more important to me than how well the seams between samples match up because I play more sustained chords than quick runs.
As for the number of samples I need, I try to use four or five in five octaves - four if the sound has a particularly long evolution that I want to capture. For four samples, I divide the memory into quarters, place the lower three samples at the three Cs centred around middle C, and the fourth around the high F or G to try and lessen the transposition problems. For five samples, I divide the available memory into fifths, place them at the five As - one sample to each octave. Those who want to hide the seams and transposition effects between samples will have to take more samples placed more closely together.
Two last adjustments are often necessary. If the sound seems to evolve more slowly at lower notes than higher ones (the envelope tracks the keyboard, or the beating is slower), I'll make the lowest sample about 20-25% longer and the higher sample that much shorter. Second, I always take a bit more sample than I think I'll need - it'll come in handy for crossfade looping (see Part II), and acts as a general safety margin.
DECISION MAKING OVER, it's time to actually start sampling. Whatever you do, keep an eye on your recording levels. Digital clipping is far nastier than you might imagine and certainly nastier than anything you'd want to hear anywhere other than with the click at the start of a drum sound. Listen to the sample after you've taken it, if it's not right, resample it - you won't get the chance later. And don't rely on editing - much as I like visual editors, I haven't been able to convincingly smooth out a clip yet.
"Digital clipping is far nastier than you might imagine and certainly nastier than anything you'd want to hear other than at the start of a drum sound."
There are several other good reasons to listen to a sample right after you've taken it. An obvious one is checking for crackling or broken cables. Another is to check the sample length and bandwidth are up to the job (or if they're doing it too well, in which case you can save some memory). Make sure that enough of the attack of the sound is present - improperly set up auto-triggering can cost you irreplaceable attack transients. Also, strangely, not every identical keystroke on synthesiser sounds the same. The oscillators or LFO's may have been out of phase at one particular moment, or a weak voice in the synth may have been triggered. In all cases, if you're not happy with it, take it again. Trust me.
Trimming at this point means getting rid of any sound that I don't intend to use. If it was a struck or plucked instrument, find where the sound fades to silence, go a few hundred samples beyond that and throw away the rest. Many samplers allow their zero-crossing detectors (see Part 1) to also be used for finding start and end points. I always trim at least the start to a zero crossing, to eliminate any unwanted click (just like you do for loops). If your sampler doesn't have this facility, set up a temporary loop point at the start of the sound, and use its zero crossing detector. Next, butt the start point up against the loop point, and throw away what's left at the start. If you foresee playing the sound in reverse, do the same to the end point - it becomes a start point when it's played backwards.
If you triggered your sample by hand, or if your auto-triggering didn't work very well, some additional trimming may be needed. In the latter case, you may find that sampling started later on a couple of the samples than on others. You can compensate for this by trimming away part of the beginnings of the other sounds, until they all sound the same at the attack. If you want to trim away the silence and noise at the start before the sound actually begins now's a good time to do it. To do this transpose the sound as far down as it will go as any silence becomes a long delay between key on and the sound starting. This is a particularly good technique to use for percussive sounds. (Beware Emulator IIs without the Attack Mod need about 5 to 15msec of silence at the start of the sound in order not to miss the attack; DPX1's up to version 1.4 also need a couple of milliseconds.)
LOOPING HAS ALREADY been covered in detail in this series, but here's where we really get stuck in. So far, I've talked about whether to sample envelopes with a sound or use the sampler's own envelopes after sampling. Aside from that, there are a few other tips that can be learned. One is slowing down the attack rate on the amplifier to smooth out any remaining clicks at the start of the sound. Another is adding punch to the sound by setting a fast attack, a moderate decay, and a sustain level of around 60-80%. This makes the attack portion of the sound louder (relative to the sustain) than it was in the original, and is very similar to using a compressor with a slow attack rate to allow the attack of a sound to punch through while attenuating the rest. Proper release times go a long way toward making a sound more believable; I've heard at least one factory piano disk that had the release set to instant off. Pianos just don't do that - it takes at least a few milliseconds for a note to die down.
These are things that can be fixed with a short-decay reverb in the mix, or by spending an extra couple of minutes listening to and tweaking the sample. But don't forget the tricks we mentioned in the first instalment about using envelopes to hide loops by continuing the evolution of the sound, and don't forget to apply the same basic envelope to all the samples that make up a sound.
AFTER THE SAMPLES are prettied up, it's time to make them work together. It's time to determine where and how to switch from one sample to another.
The first detail I worry about is clock noise. This comes in whenever a sample is transposed so far down that the sample rate is audible. In the case of a 30kHz sample transposed down an octave, there will be clock noise at 15kHz. Some samplers have filters that track well enough to hide this; most do not. This noise can be hidden by switching to the next lowest sample when the clock becomes noticeable and, if your sampler has it, routing keyboard position (tracking) to the filter cutoff to cut the offending notes (I almost always have to do this on the lowest sample).
At this point, I spend quite a bit of time deciding where the seam between samples will be. Eliminating clock noise tells you how far down a sample can be safely transposed. Next, I check how far a sample can be safely transposed up before chipmunk effects and rapidly repeating loops begin to offend, if you're lucky these points overlap, if not, more filtering will be needed to remove clock noise from the upper sample (a dull sample is less offensive than a warbling one). Then, I lower the seam between it and the next highest sample until the transition becomes smooth. Some samplers also have positional crossfade, which means the samples overlap a bit on the keyboard - this obviously helps hide seams, too.
Chances are, however, things are still not perfect. The highest notes of the lower sample are probably brighter than the low notes of the upper. Here's where the really delicate balancing act begins: try lowering the filter cutoff so that the high notes are dulled to match at the seam. Then adjust the filter tracking to compensate for the fact that the low notes are dulled too. Start at the top of the keyboard and work your way down. One danger here is that you end up so intent on filtering the seams that you filter the high end out of your samples. Either reselect the seam or give up. But remember, for most types of playing, you won't notice the seams anyway.
I'VE COVERED GETTING sounds from source (quite often a synthesiser) to usable samples. Now you can apply whatever performance parameters your sampler has to your liking - vibrato, velocity and so on. These are details that take time but may make a sample into a sound.
One other point you should be aware of is the need to keep a stack of formatted blank disks and labels handy, and save as you go - it's far better to re-cover a little ground than lose a good sample.
And that's about it - apart from a trick I have for making orchestral samples from sheet music, but I can't go telling you all my secrets now, can I?
Feature by Chris Meyer
mu:zines is the result of thousands of hours of effort, and will require many thousands more going forward to reach our goals of getting all this content online.
If you value this resource, you can support this project - it really helps!