In our first piece on sampling 'Photographing Sound', we concentrated mainly on the process of getting a sound into our sampler in the best possible state. In this article, session programmer Paul Wiffen moves onto the process of manipulating the sample to give the most useful results in playback.

The principal difference between the new generation of sampling keyboards as exemplified by the Emulator II, Mirage and the Prophet 2000, over earlier studio-based sampling devices like the AMS or early Synclavier or Fairlight systems, is that they offer a lot more in terms of post-sample manipulation. Drawing on tried and tested analogue devices as well as newly-developing digital techniques, they allow sounds to be trimmed and rearranged before performances. We will look at how the wealth of experience in analogue synthesis can be exploited in a future article, but first, as this is the order in which one would normally process sounds, let us concentrate on the digital side of things.

As we observed in the last article, there are many problems in getting a sample recorded into digital memory. Leaving aside the question of sound quality (which is first and foremost dependent upon the format of the sampler you are using), the fact of the matter is that most musical sounds can be very expensive in terms of the memory they use up, and even more so if additional data is recorded which is not required or even unused in playback. To manage the memory of our sampler efficiently we must, like governmental reformers, be constantly on the lookout for waste and ways to remove it.

STARTING SAMPLE REPLAY

The first and most obvious place to look for wasted memory is at the beginning and end of sounds. Unless you were using a perfect threshold trigger setting (to begin the sampling process) and you were lucky enough not to trigger on extraneous noises just before the sample began, then the chances are that you have recorded unwanted sound, or even silence at the beginning of the sample.

Listen carefully when you play the keyboard, or even better, sequence the sound up against a drum machine. If the sample sounds 'late' (in terms of its musical timing) then this may well be the case. Try moving the sample start point into the sample a bit and see if you notice a difference in the sound. If not then you are probably just cutting out delay in terms of unwanted memory. However, before you make any such editing permanent, check against the original start point as, when you move in bit by bit, any slight changes build up undetected but can severely alter the character of the attack.

By making a later starting point permanent without checking, you may fall into exactly the trap we avoided last month by not setting triggering thresholds too high ie. cutting off part of the attack of the sound, thereby changing its inherent character. If there is any danger of this (particularly with percussion sounds) play safe by sequencing them and syncing the sequence slightly ahead of the beat (possible with Roland's SBX-80 or a Friend Chip SRC). If you definitely want to play the sounds from a keyboard then practice playing notes a bit early. Both of the techniques shall prevent sample sounds from making the music drag.

Of course, if you have access to an on-screen sample editor (like the VES/Apple II package for the Mirage or the Sound Designer/Mac package for the Californian samplers) then it really does take the guess-work out of fixing the sample start point. If you bring the front of your sample upon screen, then you will probably see something like Figure 1.

Point 'A' shows the small sound on which the sample triggered and the'B's indicate other similarly unwanted vibrations. The required sound does not start until point 'C'.

So, now we can see this, there is nothing simpler than to mark 'C' as the start of our sample. Place the marker, which is traditionally used as the start point, on the last zero (or close to zero ie. the horizontal x-axis) point before the build-up of oscillations begins. In any editing, we always use zero-crossing points (where the waveform crosses the x-axis, representing zero sound energy), because otherwise clicks and buzzes, which are caused by sudden differences in sample readings or the gradient of the waveform slope, can spoil the sound.

Clearly, it is easy to find zero-crossings if you have a visual display, but what if this is not available? Don't despair! Many machines have an automatic computer process which locates these points of zero energy when you are looping (where this problem is most noticeable).

One, the Prophet 2000, can be made to step between zero-crossing during any of the edit functions (Start, End, and any Loop or Splice point). Whenever possible, you should make use of such facilities, as the zero crossing is often the key to undetectable editing of samples.

Samples lifted from music recordings or mediums with a lot of background noise (I told you last month not to use that old Philips cassette recorder!) can cause more trouble. If sampled with a low threshold (or none at all) they will tend to look like Figure 2a. Or, if the threshold was set too high, something more like Figure 2b. Whereas in splendid isolation, the original sound source was producing a waveform more closely approximating that of Figure 2c.

In neither case do we have an accurate attack for the sound, and marking 'A' as the start point in Figure 2a will only succeed in giving the same problem as Figure 2b has ie. a truncated unnatural attack.

One way round this problem (once you have re-sampled 2b so it looks like 'A') is to set the start point at which the sound would start if not recorded in this context-point 'B' in the diagrams. Now at least the overall attack length is accurate, and we may not notice the extremely short, inaccurate start. However, the real purist, having done this, will want to try and modify the front to make it more accurate still. Those with Waveform Drawing on their Visual Sample Editors may want to attempt to draw this correction freehand, whilst those with Analogue parameters could use the Volume Envelope Attack to trim the front into the required pointed cone shape.

ENDING SAMPLE REPLAY

After all the trickiness of setting a sample start point, setting an end point might seem simple. You just stop it when the sound you want stops, don't you? Well, whilst this is in essence the case, there are a few factors you should take into consideration. Stopping a sample replay any old place is just as likely to produce an unwanted click as starting it anywhere. Use zero-crossings again where possible to minimise such problems.

If your sound doesn't end naturally (because you sampled it from a recorded piece of music, or your sampler stopped before the sound was finished) don't despair, as all sorts of tricks can be applied to make up for this.

This is because the end of a sample is less critical. Firstly, the ear is more inclined to notice things that begin, not end (which is why we took so much care in getting the start point right), and secondly, when played from a keyboard most sounds are often stopped (by releasing a key) before they finish.

However, this doesn't mean that we should manage without accurate sound endings if we need them. Let us take the example of a snare sampled from a bit of music. Our overall sample might look like Figure 3a: 'A' would be the point in the music where this snare sound emerges from the background (be it noise or music). Mark this as the start as we saw (using a previous zero crossing). Now the snare is still dying away when another sound happens at 'B', interrupting the natural decay of the snare (or its associated reverb). Mark the end as the first zero-crossing before 'B'. If you do not have a visual editing option, move the end of the sample until you just hear a click which is the beginning of the next sound. Then move back in using as small a step as possible until the click disappears.

You have now managed to salvage the maximum possible length of sound available. The end of your sample magnified probably now looks something like Figure 3b, ending abruptly at point 'A' instead of naturally dying away to point'B'. We can reinstate this natural ending in several ways. Firstly, use an amplifier decay or release control to shut down the envelope a bit more quickly so that it ends smoothly at point 'A'. Or add a little reverb to the sound so that it dies away authentically, adding the missing shaded area between 'A' and 'B'.

SAVING MEMORY SPACE

On many sampling keyboards, there is a way to recover those areas of memory which you are not listening to ie. before your Sample Start and after your Sample End points. It is not always immediately obvious how to do this. On the Emulator II, for example, it lurks unmentioned behind the "Make Truncation Permanent? Y/N" screen. If you hit Yes then the unused portions are returned to the available 'pool' of memory. On the Prophet 2000, it is necessary to specify the sample number before you hit Execute on the Recover Memory function.

On some keyboards such economies are not possible. On the Synclavier for example, using memory wisely (like not doubling voices for chorus effects) is taken as a sign of poverty. "Not enough memory, sir? Then buy another Winchester at X grand!"

On the Ensoniq Mirage it is also not possible, but this is due to reasons of real financial economy. However, if you do find that you are only using a small amount of the memory allocated, then I suggest you delete, re-allocate less memory and then re-sample your sound. In terms of multisampling (where memory is priceless for squeezing in as many different pitch samples as possible), I cannot recommend this too highly.

However, the next topic we are about to move on to is probably the ultimate answer in terms of space saving techniques.

ECONOMIC LOOPING

Soundwave vibrations are essentially very repetitive in nature. In any sample, you are probably recording the same basic waveform (with variations in level and brightness) over and over and over again. Pretty wasteful, huh? If only we could sample this waveform less often and repeat it automatically in the sample replay, this would save loads of memory. Well, we can! Our only problem is to know where the waveforms required are and how to specify the exact repeat points.

Before we can specify actual loop points, we need to make certain that the general area we are looking in is suitable. For this purpose, we can straight away rule out the beginning of most sounds, for it is at the front that the most radical changes of timbre and volume take place. Be it the initial crack of a drum sound, the breath on a flute, the bow beginning to move or a hammer/finger on a string; the only effect of looping this portion of the sound will be to obtain a repeated, rhythmic retriggering effect. Fine, if that's what you're looking for (although the speed of repetition will change as you go up and down the keyboard - much better to retrigger the sample from a sequencer), but the big advantage of looping is to stretch short, single sounds out to the required length, not create multiple triggers: the means to do that has been around for at least 20 years.

Similarly, the end of a sound if it is decaying is not a good idea, as any change of level exhibits itself as a tremolo effect in looping. Again not wholly unpleasant, but also changing in speed with pitch. The signal-to-noise ratio is also liable to be poor towards the end of the sound.

The ideal loop area is a plateau (like the sustain portion of an ADSR) which is neither too variant in level nor harmonic content. The reason for this is that we want to use it like a sustain level of an envelope to hold the sound constant, once initial changes have taken place (in the Attack/Decay), until we release the key triggering it.

Whilst you can use the ear (with practice) to spot suitable loop areas, it is obviously easier once again to use a visual display to pinpoint likely portions of the sample. If you can, 'flick' through the various 'pages' of your sample display until you find repetition occurring. Again keep an eye out for repetition of basic waveshapes (meaning fundamental harmonic content is constant) and similar amplitude (meaning level is constant). Moving between different levels with the same waveform or different waveshapes with the same level will give you tremolo/wah-wah effects.

LOOP LENGTH

This is another tricky decision. Setting too long a loop length will exaggerate harmonics and level decays, whilst too short and you may lose the fundamental character of the sound as you fix on one particular waveform which is actually one of several that cycle around (not to mention the increasingly obvious beating which occurs).

In simple sounds with uncomplicated, unchanging waveforms (flutes, organs, etc), a short loop is often best and very economical too. What's more, you can experiment with extracting individual waveforms from a sound and using them as the basis for synthesis with conventional envelopes and filters. Aha, creativity creeping back in for a moment there (more of that in a later episode, folks)!

However, for complex sounds you will need to resort to longer sample loops, otherwise your sound will suddenly change as it enters the loop. What you need to capture is a whole cycle of harmonic changes (a succession of different waveforms which repeat naturally in the sound). This is the most challenging of all loops to create, but the most satisfying if you succeed. Don't be dismayed if the sound appears to change, as it may be part of a slower oscillation rather than the high speed oscillation of each waveform. Complex sounds, ie. ones with chorus or phasing effects, may be virtually impossible to loop (we did tell you to leave that effect off the sound source last month, remember?), but it's always worth a try - miracles do happen.

ZEROING IN

Once more, zero-crossings are the order of the day as far as looping is concerned. Find them by whatever means available (on-screen, using computer auto-looping or simple trial-and-error by ear). You will know when you don't have zero-crossings, because your loop point will be as obvious as a black cat in snow. Clicks, buzzes and other unpleasant noises will abound at the place where the loop occurs, becoming more obvious as you move up the keyboard (faster and louder as the pitch increases). Make sure if possible that both your Loop Start and End points are at zero - no use one without the other (Figure 4a/4b). Even if the waveshapes before and after the loop point are not identical, your loop will be less noticeable if you have zero-crossings at the start and end.

Clearly, with a visual system it is possible to specify loop points which are not zero-crossings. By looping between similar gradients (both going up or down at the same rate of slope) you can often create loops which are just as good. Without a visual system though you cannot find such slopes yourself.

Some machines (the Emulator II and the Prophet 2000 for example) use zero-gradients (as we call these loop points) to find ideal points automatically for backwards/forwards loops, which run from Start point to End and then reverse back to Start (instead of Start>End, Start>End ie. always forwards). These loops are life-savers for some sounds, yet completely disastrous for others. All you can do is experiment.

In essence, a loop is a digital splice of a sound back on itself and no sound is ever exactly the same for two milliseconds continuously. So don't be down-hearted if you don't find the ideal loop straight away! It takes time and patience with a bit of bluffing your way and lots of luck. If you're not a patient person then this is probably not a technique for you. But if you can keep faith, then you can from time to time astound yourself.

Try and arrange time to work on loops separate from sampling, as you don't need the pressure of "studio time is money" hanging over you until you have a lot of experience and develop a looping instinct. If you find yourself having to loop sounds in such circumstances, then do what I do: send the band and engineers off for a cup of tea (or anything else that'll keep them off your back whilst you work). You'll find things come easier without people muttering about "X pounds-an-hour" behind your back.

Monitoring on a good pair of speakers at reasonable level helps (even if it drives everyone else nuts). Poor speakers and low level often hide imperfections in your samples which then become glaringly obvious in the studio. So take care.

CREATIVE SPLICING

On to something a bit less imitative now. If we can splice sounds into a loop, then why not splice sounds to each other? It's an area for experimentation rather than recreation of existing noises and one that is a lot of fun to play with. The possibilities, as the brochures always say, are endless and though a lot of your attempts may be unusable in a traditional musical context, you can get caught up in all sorts of pretentious creative experiences. Great, isn't it?

Essentially, you can literally stick any two sounds onto each other (although you should set zero-crossings at the splice points to avoid a third sound - a click or buzz-joining in to spoil your fun), and the whackier the sounds the better, in terms of entertainment. There are no rules. Either tack a couple of sounds together for an unorthodox solo (drum/piano or guitar/flute, for example) or use up the entire memory of the machine making sound collages a la Musique Concrete. It's up to you. Why settle for piano, brass and strings (as so many people do with samplers) when you can blend all three and an anvil or clock chime along with them? Use a sampler to express your own musical personality rather than that of the guy who hit the snare on Bowie's 'Let's Dance' or the LSO strings off some CD.

With that overdone plea for less conventionality in sampling (and I mean that most sincerely folks), we come to the end of this instalment. Next time we shall look at shaping our new digitally edited sounds with our old friend VCFs and ADSRs and then move on to Multi-sampling and Overdubbing (or how to make your sampler sound like a 9ft Bosendorfer piano-well, almost). Till then, make new sounds not old!