ALIASING FILTER: (See SAMPLING RATE) When a sampling rate meets a signal higher than itself, a nasty sonic argument results called aliasing. The aliasing filter Polices the incoming signal and cuts off any frequencies above the danger point. Another filter sits at the output and extracts the very high frequency left behind by the clock which regulates the sampling gate. You don't want 20KHz whines on top of your digital spoons ensemble.

ANALOGUE TO DIGITAL CONVERTER: Yes, you can record a 'sample' with analogue electronics but you can do more with it later (eg editing) if it's in digital form. Analogue thinks in voltages, digital thinks in numbers, and numbers are easier to juggle.

A sampler actually takes many repeated mouthfuls of the sound it's listening to. The sample gate opens for a few microseconds, then closes trapping one slice of the sound in analogue memory. The ADC goes to work, analysing the information, converting it to a digital value which is stored in the next free computer memory location. The gate is cleared, and the process repeated. Much later at the end of the chain another converter returns the digitally encoded signal to analogue information which can be replayed as a recognisable sound. (Now see Sampling Rate).

CPU: Central Processing Unit. Flash title for the microprocessor at the heart of the sampler. The foreman which runs the factory.

FOURIER ANALYSIS: Expensive stuff. Those three dimensional graphs seen on Fairlight screens and resembling small mountain ranges actually show how each harmonic decays with time. They're all drawn next to each other to produce that receding landscape of peaks. This way you can analyse how the entire sound fades away and if you wish, edit every harmonic individually.

GLITCH: (See LOOPING). The unpleasant acoustic hiccup that occurs when a loop is badly joined. The aural equivalent of Paul Daniels' rich brown toupe giving way to proper dirty grey hair just around his ears. The search for the glitchless loop has been long and painful. Most machines let you step through the whole sample in fine detail picking the points at each end which will blend seamlessly (EDITING and SAMPLE START POINTS). Or there are tricks like alternative or bidirectional looping which can swing backwards and forwards across a sample instead of always cycling around it in the same direction.

KEYBOARD SPLITS: (See MULTISAMPLING) shows how many segments the keyboard can be divided into to playback different samples on different keys.

LOOPING: A sample may only last half a second. If you want it to sustain as long as your finger is on the key, you can ask it to repeat itself, or loop. On a genuine grand piano you don't hear the initial strike of the hammer over and over again. You hear it once and then the strings continue to ring on smoothly supplying the sustain. Therefore to make samplers sustain convincingly you have to isolate and repeat only the smooth, ringing part of the sample. The new generation of samplers let you start with that part of the sample containing the hammer strike and then loop with the smooth bit. It's tricky. The simpler alternative is to use only the smooth section but feed it through independent envelope generators, as on a normal synth, which supply an artificial attack.

MULTIPLE SAMPLING: With crude samplers one note is recorded (say middle C of a piano) and played back at different speeds to reproduce the whole range over several octaves. Unfortunately if you shift the pitch of a sample too far from its origin, it very rapidly sounds nothing like the original instrument — or any other instrument — at all. (Munchkinisation the great Julian Colbeck once dubbed it.) Increased memory capacity has allowed multisampling where perhaps each octave of the sampler's keyboard may have its own recording to work with. The greater the number of samples and the smaller the divisions, the better. Of course there's nothing to prevent quite different noises being mixed giving you an octave of elephants, an octave of strings, an octave of lampshades struck by Coke bottles, etc. Some devices, like Technics' PX-1 piano, now boast a separate sample for virtually every note of their 88 part keyboard.

SAMPLING RATE: (See ANALOGUE TO DIGITAL) refers to how often this listening' process repeats and is expressed in KHz (20KHz = 20,000 'listens' per second). The faster the rate, the more information is gathered, and the higher the fidelity of the result. In particular, the more top end is preserved.

Because how can a 10KHz sampler capture a 20KHz signal. You don't use a pushbike to pace a Ferrari. Many other problems conspire and in practice, the maximum frequency to survive in the sample (and that includes the harmonics which make up the brightness of your signal) is just under half that of the sampling rate. The new Roland S-50 has a maximum sampling rate of 30KHz offering a frequency range of 20Hz to 13KHz. In comparison the Casio SK-1 reviewed last month was 9.3KHz.

SAMPLE TIME: There's only a limited amount of memory, but it can be used in different ways. If a rate of 30KHz produces a one second sample, then cutting it to 15KHz will allow two seconds, but at half the maximum frequency response — you lose top end. The machine has less to memorise, so more space to do it in.

WORD MEMORY: (See ANALOGUE TO DIGITAL). Each of those slices is thought of as digital word — a fashionable tech-phrase to drop into conversation but a useful method of describing how much sampling space you can fit on a storage disc.

12BIT/16BIT: Generally the bigger the word (the more letters in it, if you like) the better. Crudely put, 16BIT technology (a measure of the power of the integrated circuits used) is a bigger piece of paper than 12BIT, and allows you to write a longer and more detailed word before turning the page. As we know, more informations means higher-fi, and 16BIT samplers should exhibit improved quality if the rest of the circuitry is up to scratch.