Most samplers on the market today are provided with individual voice outputs to allow separate mixing and external processing of their sounds. However, the undeniable advantages this gives are not without their price. Paul Wiffen expounds.

The provision of separate outputs on electronic instruments is one of the more commendable responses from manufacturers to pleas from end-users, particularly those in recording studios, for the ability to process the huge variety of sounds that modern samplers and synths can simultaneously produce, independently of each other.

Of course, it was on digital drum machines that the need for this was first felt, and it was a need manufacturers were happy to fulfil as the machines were at first expensive enough for the cost in extra hardware to be but a small percentage of the overall cost of the machine. In addition, as we were all so thrilled to be hearing authentic drum sounds from a box for the first time, we didn't notice until years later that successive strikes of the same sound cut off the previous notes.

As the price of the digital technology began to fall, manufacturers (like Sequential in the Drumtraks and Emu in the Drumulator) began to economise by offering fewer outputs than available sounds and having the less standard drum sounds share outputs. After all, the hardware for the outputs was now costing them more, even though the chips in the machines were but a fraction of their former price.

It was then that we began to notice that the low tom stopped the high tom from ringing and that we couldn't play all the additional percussion sounds at once. Still, a dash of reverb in the studio cured the first problem and, really, who cared that you couldn't get cabasa and wood block together?

Of course, when you were talking about the really cheap drum machines like the Roland TR606 (which really needed individual processing), then there was nothing for it but to settle for a single audio output. This, as we know, spawned a whole series of articles in magazines like this about how enterprising electro-musicians could fit their own separate outputs and save themselves the manufacturer's, distributor's and dealer's mark-ups on half-a-dozen jack sockets. This tradition has continued and only the other day I saw a TR505 which had some rather unevenly spaced separate outputs on the back. [The new TR626 comes with separate outs as standard. See review this issue - Ed.]

So, by and large, whether we bought our separate outputs with the machine or fitted them ourselves, we were happy. We could take all the top end off the toms and phase them whilst brightening up the snare and sticking an 8ft reverb effect on it.

But there were those who noticed that a new note cut off the old one, especially when the sample duration of sounds began to grow longer and longer (well, it's bound to be more obvious, innit?). Recently, to please even these perfectionists, the more upmarket machines (Studio 440, Dynacord ADD-One, Simmons SDX) have featured a 'cycling of voices' to allow long snares, cymbals and other sounds which cause an unnatural effect if cut off by retriggering, to continue to die away while a new voice is assigned to that sound as it is retriggered. Call me fussy if you like, but I don't really find this terribly authentic. Each strike in a snare drum roll doesn't die away as if it were on its own, because the next strike will mute it somewhat (although not as much as retriggering the sample from the start). With this 'cycling' system, fast retriggers tend to pile up in an unnatural 'layered' sound that seems, to my ears at least, more like half-a-dozen drummers hitting half-a-dozen snares slightly out of time with each other. I think that a smidgeon of reverb does the job much more elegantly.

"So there we were, perfectly content with our dual layer/split keyboards, when Sequential suddenly threw multitimbral keyboards at us."

Still, everyone seems pretty much in agreement that separate outputs on drum machines is a good thing and the closest thing to a disadvantage you can find is that you always seem to have one less audio cable or channel on the mixing desk free than the number of sounds you want to use! Nevertheless, it's a lot less bother than miking up an acoustic drum kit, uses no more channels, and gives you a usable sound every time (not to mention the fact that some drummers still think Time is a magazine!).

However, back in their R & D departments, the equipment manufacturers were about to spring another surprise on us. We were all used to synths which could make two different sounds at the same time, like the Prophet 10, OBXa, Jupiter 8, and so on. And they usually had separate outputs for the lower and upper keyboard sounds, even though half the time we just layered the sounds and monitored in mono - especially as random (and radical) panning effects would result if you switched back to one sound at a time while still hooked up in stereo: melody lines particularly tend to lose their sense of flow when notes jump from hard left to hard right at random! Incidentally, this is still a problem on machines like the Prophet 2000 when you monitor in stereo. Thank goodness for the 'proper' stereo of the Prophet VS, Matrix 12 and Emax.

Where was I (after I've got that one off my chest)? Oh yes... So there we were, perfectly content with our dual layer/split keyboards, when Sequential suddenly threw multitimbral keyboards at us. Well, no sooner had we worked out what it meant (several Latin dictionaries later) and realised what a good idea it was ("You mean I can get six different sounds at once?") than, like spoilt children, we started grumbling.

We liked having lots of different patches at the same time, but did they all have to be monophonic? Well no, but you had to settle for fewer sounds if you wanted to play more than one note at a time with each sound. We learned the hard way that, however flexible the synth, each note you play requires a voice of its own and every time you increased the polyphony by one note, that was one less sound you could have.

Of course, the rich and famous (read Fairlight and Synclavier owners) could have told us this harsh fact of life. They'd had multitimbrality for years (although they'd never heard the word before) and were used to seeing their expensive voices disappear as soon as they programmed polyphonic parts into their sequencers. In fact, many a Fairlight Series II owner would have killed for the ease with which Sequential's multitimbral voices could be played polyphonically; they had to treat a three-note polyphonic part as three monophonic lines.

"The trouble is that sampling just cries out for individual outputs."

So we soon got spoilt by multitimbral use and then we started to complain about the lack of separate outputs (on the Sequential Six-Trak, for example). Which was a bit unfair, seeing as it only cost £800 when released and you still can't get a keyboard with separate outputs for that money now (four years later), even with the endless downward spiral of the cost of technology. "Useless!" was the cry, and it was only silenced by the arrival of the Multi-Trak, which had the much clamoured after separate outs.

Then, of course, we learned another hard lesson about polyphony and individual outs. When using polyphony, you had to take the outputs for the voices playing the same sound and sub-mix them together to hear the polyphonic part, before you could begin processing them with effects. Fairlight owners could have told us that too.

Up until this point, however, it was only Sequential among the 'budget' manufacturers who were running with the multitimbral torch. If you stayed away from Sequential's Trak series of keyboards, then you would have remained blissfully ignorant of the joys of multitimbrality - and the attendant need for individual outputs - until samplers came along.

The expensive sampling systems, up to and including the Emulator II, had always had separate outs - users expected it at those prices. It was the same old story as with the expensive drum machines: the cost of the necessary hardware (sockets, etc) was insignificant in comparison to the digital innards.

So it was the first two 'affordable' samplers, the Ensoniq Mirage and the Prophet 2000, which bore the brunt of the public's wrath at the lack of separate outputs. Never mind the fact that both machines cost a fraction of previous samplers; never mind that the Prophet 2000 sounded better than any other sampler before it (except the monophonic Synclavier).

"It is primarily the need to process samples through a standard VCF... which prevents dynamic voice allocation from being usable with separate outputs."

The trouble is that sampling just cries out for individual outputs. To begin with, sampling lends itself to multitimbrality because of the way the sampled waveforms are read out of memory. Instead of an endlessly repeating wave cycle, the sample is stored in memory in the form of a long stream of numbers. To change the sound coming out, you don't have to reprogramme the oscillators and filters, just read some different numbers from a different area of the RAM. And because you can access different parts of RAM at the same time (that's why they call it Random Access Memory), there's no reason why you can't have as many samples as you have voices at the same time.

Couple this with the fact that drums are the most easy and convenient things to sample (why do you think we had digital drums before anything else?) and you'll understand why it didn't take anyone who bought a sampler too long to start wanting individual outputs. Especially when they were using something as wonderful as the Mode 4 (Mono Mode) application of the Prophet 2000, where you could get up to 16 sounds being sequenced at the same time, and although you knew you only had eight voices to play with, it sounded like more because of the way the Dynamic Allocation feature switched them around. As long as you didn't have more than eight notes playing at any one time, you could have loads of different sounds interweaving and simulate a dozen or more musicians playing together. All you needed were the vital separate outputs. At least, till you got them! Then the old problems of monophony and number of voices reared their heads.

Take the Akai S900, for instance: if you used the Mix Out socket, the S900 would do all the wonderful polyphonic Mode 4 stuff that the Prophet did, but with 32 samples across 16 MIDI channels. But as soon as you started assigning voices to the S900's eight individual outputs, you were back to the old eight monophonic signals limitation. And I remember that sinking feeling the first time I played one of my Mode 4 sequences on a Prophet 2002 Plus with the separate outputs assigned: the same thing - eight monophonic lines is the best you can get.

If you are just using a sampler for playing drum sounds, then eight monophonic parts is probably all you need. Similarly with bass and melody lines, you will find that these machines do the job. It's only when you come to create your three-note poly string parts and four-note brass chords that you find the limitations.

Actually, on the Akai S900 you can set up one four-note polyphonic part along with four monophonic (ie. one-note) parts. This is because the first four individual outputs are grouped together for the Left output, while outputs 5 to 8 are grouped together for the Right. So by assigning four samples which you only need monophonically (eg. bass drum, snare, hi-hat and bass line) to outputs 1 to 4, and a four-note brass or string part to the Right output (number 10), you can achieve the desired result. Alternatively, you could make outputs 5, 6, 7, and 8 the monophonic ones and monitor your polyphonic part via the Left output (number 9).

There will undoubtedly be two sets of readers of this article who won't see what the problem is: those who own Synclaviers (are there any problems enough money can't solve?) and Roland S-50 owners. Both of these machines are able to assign voices polyphonically to separate outputs. This is principally because, unlike all the other samplers we have mentioned so far, neither of these models has analogue filters. It is primarily the need to process samples through a standard voltage controlled filter, or VCF, (to remove digital aliasing or to add synthesizer processing) which prevents dynamic voice allocation from being usable with separate outputs. Once you get rid of analogue filters, either by not incorporating them in your design like New England Digital (Synclavier) or by simulating them digitally like Roland (S-50), then you can group the voices to whichever individual outputs you like.

The moral is that if you need polyphony with your separate outputs and you can't afford a Synclavier (not everyone can, you know!), then the Roland S-50 sampler looks like your best bet at present. But if you still consider the four separate outputs of the S-50 a bit limiting, then you don't have long to wait for the rack-mount S-550 with its eight separate outputs, one megaword of sample memory and dynamic filtering (via the Time Varying Filters developed for the Roland D-50 synth). It seems that as far as separate outputs is concerned, you can have your cake and eat it after all!

Paul Wiffen is a musician/programmer who has worked for Paul McCartney and Geoff Downes, amongst others. He is currently assisting Stevie Wonder on his mammoth world tour.