Magazine Archive

Home -> Magazines -> Issues -> Articles in this issue -> View

Making the Most of... (Part 10)

Sampling

This month we focus on the hows, whys and whats of sampling.


The ins and outs (or inputs and outputs) of sound sampling.

Powertran MCS1

To obtain the best results from any process, you need to know a little about how it works, and the same is true of sampling. With a little understanding and a bit of common sense, you can get first class results from even quite modest equipment.

There's nothing new about sound sampling. Tape recorders have been doing it for years, but when we talk about sound sampling in the currently accepted sense, we usually mean digital sampling. Those who know how the mechanics of sampling work take the whole thing very much for granted but for someone looking at it for the first time, it can seem more than a trifle daunting. In terms of what a sampler can do, there are really only two types in what I would call the affordable bracket: those that simply sample a short sound and replay it again on receipt of a suitable trigger pulse, and those that may be controlled in a musical manner via a keyboard. Both types have a limited storage time, usually only a few seconds, and this is dictated by the amount of memory that the machine has. As memory tends to be expensive, the manufacturer has to come to some sort of compromise as to how to use the memory most efficiently.

There are two main factors that determine how much memory is needed; the length of sample (or sampling time), and the bandwidth. As you might guess, the longer the sound you want to sample, the more memory you will need (given that all other factors remain the same), but bandwidth also eats up memory so the designer has to juggle these two parameters against the available memory in order to come up with a cost effective design that offers acceptable sound quality and also gives a usefully long sampling time.

Aliasing



To understand why sampling time and bandwidth are to some extent dictated by the amount of memory a machine has, we need to examine the theory of sampling, so skip this bit if you aren't interested in the technicalities. Sampling is the electronic equivalent of cutting an electrical waveform into slices, and by measuring the height of each slice in terms of volts, the information can be digitised and stored. When these numbers are retrieved at some later time, they can be fed back into a digital-to-analogue converter which will turn them back into slices of voltage thus recreating the original waveform. However, to accurately reconstruct a waveform in this way you need to have at least two slices for each cycle of the highest harmonic present, otherwise you get a nasty effect known as aliasing which is in effect an audible beat frequency between the input signal and the sampling frequency (the rate at which the input sound is sliced). In practice this sounds like a metallic or whistly overtone which is not musically related to the input in any way. To avoid feeding illegal frequencies into the sampler then, the circuitry must contain a filter to remove any high frequencies that might cause trouble, and as filters are not perfect devices, they have to be set a little lower than the theoretical frequency to make sure that nothing untoward gets past them. In practice this leaves us sampling at nearer three times the highest input frequency rather than twice it and this figure is typical of most commercial samplers.

Another filter similar to the input filter is fitted at the output, again to prevent aliasing due to high frequency components present in the reconstituted signal. A sophisticated design might have a filter that automatically varies as you adjust the sampling time thus giving you the widest possible bandwidth for all settings.

Starters Orders



Some samplers have a built-in sound operated trigger which starts the loading process as soon as a signal is detected at the input, whilst others require you to push a switch at the same instant as the sound starts. Both methods have their drawbacks.

With the manual switch system, you need to co-ordinate the start of the sound and the start of the sample very accurately, otherwise you may either chop off part of the sound or record a short pause before it starts. On a machine with editing, these pauses can be removed, but on a budget unit you may well not have much in the way of available editing features. The other problem, (chopping off the beginning of the sound) is more serious because there's no way of restoring the missing part of the sample (though many percussive sounds are quite acceptable with a small part of their leading edges chopped off). Indeed you can do this creatively using edit controls, if you have them, and it's nice to be able to experiment on percussive sounds in this way.

The sound operated trigger is far more convenient and works well in most cases but there are pitfalls. Firstly, any sound over a certain threshold triggers the recording and so an intake of breath before a sung word or phrase could set it off by mistake. You can get around this by pressing the Ready To Record button just before singing but after drawing breath. The other problem of course is that of chopping off the leading portion of a sound that has a slow attack. If you're sampling from tape and have some form of editing facility, then you could try the following; thread the tape onto the machine backwards and then play the sound. At the end of the sound (really the beginning) record a click or a beat from a rhythm machine and then play the tape the right way round. Now the click will be at the start of the sound and this click will trigger the sampler. Once the sound has been loaded into the sampler, the click may be removed using the edit controls and you will be left with only the original sound with all its attack characteristics preserved.

Trigger Delay



Korg SDD2000

One figure that most manufacturers seem reluctant to publish is the trigger delay time which is the time it takes for the sampler to realise that it has received a trigger pulse and to start to play back the sample. On some machines this can be quite a few milliseconds and this can affect the sense of timing of a piece of music, especially if it's part of the drum track that you're sampling. An easy way to check this is to use a good drum machine with an external trigger output to drive the sampler and to sample one of the drum voices (a technique pioneered by our very own John Harris). A rim shot or similar short sound is ideal. If you then arrange for the drum machine to play that voice at the same time as it is triggering the sampler to play the same voice, you can add the two signals and listen for any delay. In the worst cases the two sounds will be separate and give a flam effect whilst a better unit may only produce a slightly flangey or tunnely effect to the sound, due to the very short delay involved. If you're triggering the sampler from clicks on tape or from a drum machine synced to tape, you can reverse the tape and copy the click or code track via a DDL set to the appropriate short delay time and re-record onto a spare track. When the tape is once more played the correct way round, the timing track will appear slightly earlier and compensate for the delay in the sampler.

Looping



As previously stated, a budget sampler is going to have a fairly limited storage time and so any attempt to sample a long sound is going to result in the end of the sound being chopped off, and this invariably sounds wrong. If the unit has no looping facility, then you really have to try to engineer the sounds to fade to nothing by the time the sampling period runs out, but if you're lucky enough to have looping, then you can really start to have fun. With looping, what you are doing is allowing the sound to die away for a time and then repeating a short part of the sound round and round to produce a continuous note. Sadly things are never as simple as they seem and it's difficult to find suitable loop points that don't betray their presence by a series of glitches or clicks at the point where the ends of the loop join up. These clicks may be caused by a difference in level between the start of the loop and its end but this can be avoided by making the loop very short. More serious is the fact that the waveform may be continuously changing and that even if it weren't, it's still difficult to get the ends of the loop to join at the same part of the waveform. Normally the edit controls are adjusted whilst listening to the result and the point of least glitch is evident when you eventually find it, but some of the more sophisticated samplers actually help you by joining up the loop ends at the points where the waveform crosses the zero volt line. Either way you can usually hear the effects of looping when a sound is played in isolation but when it's mixed into the track or effected in some way, it usually all but disappears.

So there we have the first compromise. The thicker the slices, the longer storage time you can get from a given amount of memory, but the frequency response will suffer accordingly. For most purposes an upper frequency response of 10kHz would sound bright enough but anything less will tend to sound noticeably dull when compared to the original. For a sample that is as bright as the original, you should aim for a 16kHz bandwidth or higher, but only expensive studio machines are likely to offer this kind of specification.



"If the unit has no looping facility, then you really have to try to engineer the sounds to fade to nothing by the time the sampling period runs out..."


How Many Bits?



When we refer back to our sliced loaf model of a sampled waveform, it's easy to see that there's another factor that affects sound quality. If we measure the height of each slice and then store it in memory as a number, the measurement accuracy or resolution is very important. An 8-bit system only gives us 256 levels of resolution and though this may seem like a lot, the fact that input signal may contain both loud and soft sounds means that the slices at low signal levels may only be a few levels high and so very large errors are incurred. These errors show up as noise and this is given the endearing title of quantisation noise.

One way of improving the performance of an 8-bit design is to use a companding system which acts rather like DBX noise reduction in that it compresses signals on record and expands them on playback, but the method is very different. Rather than using compressor and expander circuits, the trick is performed at the point where the analogue slices are measured and digitised: in the analogue to digital converter. By making the measurement steps non-linear, it can be arranged so that the levels are closer together at low signal levels than they are at high ones, so the lower amplitude signals get more than their fair share of steps. This has the effect of reducing the quantisation noise at low signal levels, and at high signal levels the noise is masked by the signal itself. Of course the output converter circuits are designed to compensate for the non-linear law so the output still sounds correct. Better resolution can be achieved with a 12-bit system but many people actually prefer the slightly artificial quality imparted by 8-bit sampling which is why the Fairlight CMI is still so popular.

Because a 12-bit number takes up more space in memory than an 8-bit number, we again need to use up more memory to improve the quality.

To summarise then; there are three main factors which set limits on the sound quality of the end result and all require the use of more memory if the quality is to be improved; bandwidth, maximum sample time and sampling resolution.

Getting Loaded



The first job is of course loading the sound into the sampler and you may be surprised to know that you may well get better results by sampling a recorded sound than by sampling a live one. With a recorded sound you have the option of repeatedly running through it to allow you to set up the right recording level on the sampler, but with a live sound, every version will be slightly different.

Akai S612

As budget samplers are a compromise in terms of noise and resolution, it's essential to get as much level into them as possible without distortion if good results are to be expected. The effect of overloading a sampler depends very much on its circuitry, and if the analogue input stage limits first, you may be lucky and end up with a fairly subtle distortion. If however the digital circuitry saturates first, there is no safety net; you simply run out of numbers and the resulting distortion can sound very nasty indeed. When you sample a previously recorded sound, you have the added benefit of tape compression which reduces the dynamic range of the signal without making it sound unpleasant.

Some of the slightly more up-market samplers such as the Bel BD-320 offer relatively long sampling times, and so rhythmic patterns or sequences of notes may be stored and you can even sample whole lines from songs. This is handy when you're recording a difficult chorus as you only need to get it right once and then you can drop it into the mix as required.

Finally, most samplers, even the very cheap ones, give you the facility to overdub sounds so that a sample may be built up to consist of several layers. This is very effective for creating rich choral vocals and if you do have enough memory to store whole lines of vocal, you can add the harmonies at this point to save on tape tracks.

It you do loop a sound, presumably the note stays at the same level until a new note is played, so what can be done? Well, it depends on the design of the sampler. Some will continue with the natural decay of the note once a key is released (assuming keyboard control) whilst others will simply cut off the sound; some give you a choice but at the budget end of the market, this area is fairly rudimentary.



"You can disguise the end of a sample to some extent by adding reverb but there's no substitute for a long sample time if you want to do this job properly."


Envelope



If you're using a budget monophonic sampler that may be controlled from a CV/gate type keyboard, you have an ideal opportunity to modify the sound even further. Most monosynths have an audio input which bypasses the oscillators and goes straight into the filter and envelope sections. If the output of the sampler is fed into this input, the sound may be given a filter sweep and a totally new envelope, and you can even mix in the sound of the synth's oscillators to boot if you want a fatter sound. This also has the advantage of applying the envelope to any noise inherent in the sample, resulting in a subjectively cleaner sound. Don't forget though that you will need to tune the sample pitch to match the synth's pitch.

If your sampler has no musical pitch control facility, then you will probably find it most useful for sampling percussive sounds and so the end of the envelope should not give you cause for concern as these sounds are generally shorter than the sampling time. The exception here is cymbals which require several seconds of sampling at a good bandwidth to sound realistic. One way of cheating is to loop part of the decay of the cymbal but this is only likely to sound convincing on very sophisticated machines. You can disguise the end of a sample to some extent by adding reverb but there's no substitute for along sample time if you want to do this job properly.

In general, percussive sounds are driven from the trigger output of a suitable drum machine, though some will trigger directly from a microphone, giving you the opportunity to use conventional drums to initiate the sounds. In this case you will almost certainly need to use a noise gate to ensure that only the required drum triggers the system.

Pitch Considerations



When a sound is sampled and then changed in pitch, this is accomplished by speeding up or slowing down the sample and if you've tried this trick with a tape recorder, you'll know that you can only go so far before the sound becomes distinctly odd. In practice a sound remains usable for around two octaves but if you really want to create a natural effect, a range of four or five semitones might be nearer the mark. However, one of the beauties of sampling is that a fairly ordinary sound might be quite exciting when drastically changed in pitch, so don't limit yourself to natural sounds unless you have a good reason for doing so.

Digitech PDS2000

Up-market machines such as the Fairlight get round this limitation by letting you multi-sample. This is a way of using several samples recorded at different pitches to cover the keyboard such that no sample is transposed far out of its natural range before the next one takes over. Even lower priced machines such as the Ensoniq Mirage and the Prophet 2000 feature multi-sampling but none of the budget rack mounting units have this facility... yet.

MIDI



You can't go far these days without running into MIDI and this is also true of sampling. With the advent of MIDI keyboards, the floodgates have been opened for a whole host of MIDI-compatible equipment and, as a sampler needs a keyboard or other source of control to make musical sense, MIDI is a logical step.

The Akai S612 features MIDI control and it is the least expensive sampler to give you true polyphonic capability. With this machine you can play up to six notes simultaneously and, via MIDI, it is possible to implement pitch bend and modulation as well as playing dynamics. Considering that this little unit costs under £1000 and comes complete with disk drive for the storage of samples, it certainly makes a lot of sense. This brings us conveniently onto the topic of sound storage.

Dumping



It's all very well loading up unique samples, but what happens when you switch the sampler off or if you want to use a different sound? Almost invariably you lose your sample.

With the cheaper units you only have the option of saving the sound on tape in the conventional way and then re-sampling it whenever you need to use it. This is inconvenient to say the least, especially if you've spent hours locating suitable looping points. That's where the disk unit comes in.



"...many effects contain a frequency component so when the pitch is changed by speeding up or slowing down the sample, the effect will change too."


Disk drives have been used for a long time in the computer field to store programs or data and the stored data is more or less permanent if you don't damage the disk or overwrite its contents. The Akai machine uses miniature disks to store its sounds and, though you can only store one sound on each disk, the system also memorises edit and loop points which can save a lot of time and frustration and indeed, makes the unit a practical proposition for live use.

All the serious machines seem to be following this approach but it's only a matter of time before the protocol for sample transference via MIDI is ironed out (indeed, most of the work is already complete) and then you may be able to store your treasured samples on rampacks as well as on disk or analogue tape, not to mention the disk drive of your home computer (with a suitable MIDI interface).

What to Sample



No I don't want to tell you what sounds to sample but rather in what form to sample them. However tempting the library samples might be, it's always best to create your own samples where possible, because library samples have a habit of turning up all over the place and become readily recognisable.

Ensoniq Mirage

In virtually all cases, it's best to sample the sound in a dry, untreated form and there are several good reasons for this.

Firstly, as you might guess, echo or reverb will lengthen a sound and take up more memory with the probable result that you'll run out of memory before the sound ends. More important though is the unarguable fact that it's easier to put an effect onto a dry sample than it is to remove an effect from a treated one.

Yet another consideration is that many effects contain a frequency component so when the pitch is changed by speeding up or slowing down the sample, the effect will change too. An example of this would be a chorus unit, the sweep speed of which would be different with every note. Likewise reverb would have a different decay time for each note, and any simulated room resonance would also change so that every note would sound as though it were played in a different sized room. Anyway, I'm sure you get the drift: use no effects with the possible exception of compression which might allow you to make better use of the available headroom.

It's also worth pointing out that you should use a decent mic when sampling, as there's no point in reducing the frequency response any more than you have to by compromising in this area. Samples taken from tape are usually subjected to a little tape compression so you may get better sounding results by doing the original recording with the meters going well into the red but not far enough to cause audible distortion. On the subject of recorders, it makes sense to use something decent and if you're using a cassette deck it should have noise reduction. If you do use a bad recorder, the results are liable to be both dull and noisy.

Some samplers offer you the option of varying the frequency response and bandwidth manually so that you can reach your own compromise between sound quality and sampling time, but don't reduce the bandwidth too much. There's a common misconception that bassy instruments such as the bass guitar and the bass drum don't need to contain much in the way of high frequencies but that is quite wrong. You can demonstrate this for yourself by experimenting with a recording of a bass drum and the EQ controls on your mixer. A bass drum with no top end will sound muddy and lack both impact and definition as will a bass guitar. True, you may be able to compromise in these areas with fewer problems than in others but be careful.

If you do need a long sample time and are forced to compromise the bandwidth by more than you'd normally consider reasonable, you can use the old faithful psychoacoustic enhancer to re-create the missing upper harmonics. I use my Scintillator for this purpose and this trick is also valid when applied to low bandwidth DDLs. Remember though, that all these enhancers emphasise the noise that is part of the input so I would recommend the output of the enhancer be fed through the synth envelope shaper to ensure a clean sound with the noise gradually fading out at the end of a note rather than being abruptly cut off.

The noise performance of any sampler can be improved to some extent by boosting the top end of the signal fed into it. If you then remove an equivalent amount of top on playback, the sound should be the same but will contain less high frequency noise. This is the same system as employed in many recorders and effects units and goes by the name of pre- and de-emphasis. The only time that this trick fails is when the sound is already rich in high frequencies in which case the boosted treble will cause the sampler to overload earlier than it otherwise would do. This leaves you no better off than when you started.

Lifelike?



Don't despair if your samples don't all turn out sounding like a Compact Disc recording of real life because many of sampled sounds on record are virtually unrecognisable but musically viable nonetheless. That line on Kate Bush's 'Running Up That Hill' for example... just what sound was that when it started life? What the Fairlight CMI can do by merging waveforms, you can do by distortion or pitch changing or by plain bad sampling, but if the end result is pleasing, then use it... after all, isn't that what it's all about?


Series - "Making the Most of..."

Read the next part in this series:


All parts in this series:

Part 1 | Part 2 | Part 3 | Part 4 | Part 5 | Part 6 | Part 7 | Part 8 | Part 9 | Part 10 (Viewing) | Part 11 | Part 12 | Part 13 | Part 14


More with this topic


Browse by Topic:

Sampling



Previous Article in this issue

Rock Around the Clock

Next article in this issue

Susstudio


Home & Studio Recording - Copyright: Music Maker Publications (UK), Future Publishing.

 

Home & Studio Recording - Feb 1986

Donated & scanned by: Mike Gorman

Topic:

Sampling


Series:

Making the Most of...

Part 1 | Part 2 | Part 3 | Part 4 | Part 5 | Part 6 | Part 7 | Part 8 | Part 9 | Part 10 (Viewing) | Part 11 | Part 12 | Part 13 | Part 14


Feature by Paul White

Previous article in this issue:

> Rock Around the Clock

Next article in this issue:

> Susstudio


Help Support The Things You Love

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!

Donations for August 2022
Issues donated this month: 0

New issues that have been donated or scanned for us this month.

Funds donated this month: £136.00

All donations and support are gratefully appreciated - thank you.


Magazines Needed - Can You Help?

Do you have any of these magazine issues?

> See all issues we need

If so, and you can donate, lend or scan them to help complete our archive, please get in touch via the Contribute page - thanks!

If you're enjoying the site, please consider supporting me to help build this archive...

...with a one time Donation, or a recurring Donation of just £2 a month. It really helps - thank you!
muzines_logo_02

Small Print

Terms of usePrivacy