The Sound On Sound Guide To Samplers
Everything you ever wanted to know about samplers... 48 models compared! Paul Gilby surveys the current sampler market, exposes the 'spec wars', dispels the 16-bit myth and explains in simple terms how sampling really works, before telling you what to look out for when buying secondhand.
PAUL GILBY surveys the current sampler market, cuts through the 'spec' jungle, dispels the 16-bit myth and explains in simple terms how sampling really works, before telling you what to look for when buying secondhand.
This Guide is the most comprehensive survey of its kind ever published in the UK. Within the article we have covered 48 different models of sampler in the form of keyboards, rack-mounts, drum machines and computer-based systems. We have not included the various outboard equipment such as digital delays and multi-effects units which often have a limited sampling feature built-in. The accompanying table of features has been compiled from many sources and includes data gleaned from product brochures as well as telephone calls to many of the companies listed.
Whilst compiling the data, it became very apparent that many inaccurate figures are being quoted in everyday conversation and that even amongst manufacturers' own literature, the specifications of some instruments vary. This all leads to a state of confusion for anyone who is trying to decide which sampler to buy. We hope this Guide helps you make a more informed decision.
The Guide is divided into two sections. One section contains the main table which lists all 48 instruments and their features along with an explanation of how to read the table and a series of extended product remarks. The other section covers various aspects of sampling technology and offers advice on buying secondhand instruments. You can read the article in any order and it should still make sense.
A sampler is a digital recorder which captures a few brief moments of sound and allows you to play it back either as you originally heard it, or in some edited form. A compact disc player can be considered a 'playback-only' sampler.
The concept of sampling and the manipulation of sound is really quite old. During the late 1940s when tape recorders were first introduced, composers were recording sounds onto tape, cutting the tape up and re-arranging it into a new order before playing it back. This idea was developed into quite an art form and became known as musique concrete - with techniques which moved the start of a recorded sound to the end; middle sections of a sound were re-recorded and spliced together to lengthen the overall duration of the sound; tape was joined into a loop so that a sound would play forever, etc. It was, however, a manual and labour-intensive task but these were the seeds of the digital sampling techniques to come.
Around the end of the 1960s, an instrument called the Mellotron was developed. This enabled sounds recorded on tape to be played back by pressing a key on a piano-style keyboard. It was very limited and mechanically unreliable, but it did allow musicians to use sounds previously unavailable. You could, for example, have the sound of a whole choir of voices without having thirty or more people in the recording studio or on stage.
The Mellotron remained the only instrument to offer this type of sampled sound for musical expression for almost a decade.
Musical instruments which allow you to sample your own sounds haven't been generally available for very long. True, the Fairlight and Synclavier have been in existence for nearly ten years but they are hardly affordable to the majority of musicians. It was E-mu Systems, with the introduction of the Emulator, who first brought the price of sampling instruments down to a more accessible level. Later, Ensoniq launched the Mirage at a price which really brought sampling into the reach of serious and semi-pro musicians, then Akai followed with the first sampler (and the first without a keyboard) under £1000, the S612.
At this time, the Fairlight Series II still dominated the professional recording studios and was the de-facto standard on many chart singles. Technically, it started to fall behind as its once state-of-the-art 8-bit samples began sounding a little 'muddy' when compared to those of the newer instruments. It did, however, continue to offer its own charm and warmth. For Fairlight, the time had come to move on.
The launch of the Emulator II and, later, the Prophet 2000 gave professional users a lower cost and better sounding sampler. The Fairlight Series III was then announced.
As all of this was going on, the Synclavier continued to evolve and its creators. New England Digital, were talking of increasing the sample time capability of the system. Eventually, they coined the phrase direct-to-disk to explain their new approach to sampling. Sounds were not just digitised into the machine's memory, as that limited the maximum length of the sample you could take. Instead, the sound was sampled directly onto hard disk and resulted in hours, not minutes, of sample time. Then they pioneered the multitrack version with up to 16-tracks. The 'tapeless' recorder was born and sampling moved on from being just about capturing small snatches of sound like a snare drum beat, to capturing long continuous sounds, whole verses, even complete 'performances'.
We are now entering an age where the creative possibilities offered by sampling technology are greater than any previous instrument technology available to the musician. One thing is certain. The important pioneering work of the Fairlight and Synclavier systems has always filtered down to the mass consumer market. With that in mind, the future of sampling looks like it's going to continue to be very exciting!
A sound exists as a movement of air pressure. If you want to keep a record of that sound, you need to place a microphone in front of it. This will convert the sound pressure into an electrical voltage which varies in time, in direct response to the movement of the air. You can store this on a tape recorder by recording the waveform of the sound on a piece of magnetic tape. You can then play it back and listen to the sound again. If you mark the tape at the points where the sound starts and finishes, you can cut the tape at those points and hold it in your hands. If the tape recorder transports the tape at a speed of 15 inches per second and the sound you recorded lasted for one second, you would have a piece of tape 15 inches long - no surprises there!
Once you have this piece of tape you can join the two ends together and make a loop. You can then use a tape recorder to replay this loop and the sound on the tape will repeat itself time after time. However, if you wanted to hear a part of the sound and not the rest, you would have to cut the tape again and you would lose what you previously had. If you wanted to hear the tape play from a certain point every time, it would be very difficult to make this happen. Enter the age of digital technology...
With the advent of microprocessors came the means to record sound not onto tape but into computer memory. This meant that it existed as numbers stored inside a microchip and not as a waveform on magnetic tape. You couldn't pick it up and measure it anymore, or cut it into pieces, but engineers soon discovered you could do far more things with it than was ever dreamt of before.
Recording sound in a digital form works rather like cinema film. You take a number of individual pictures (or samples) of live action and store them as unique events in a serial form, ie. one sample after another. To produce a moving action film, 24 separate samples of the live action are photographed every second. You could therefore say that the film has a sampling frequency of 24 pictures per second. If you look at a stationary piece of movie film you can see each individual picture, or 'frame' as it is known. When the film is played back, the speed of 24 frames per second is sufficient to make your eyes/brain think you are witnessing live action - reality.
Digital sound works in the same way, except that you need more than 24 samples per second and the information is stored as a number in a computer's memory rather than a picture on film. The question is, how many samples per second do we need to take in order to convince our ears that we have captured the sound faithfully? Well, the answer is really dependent on how accurate you want the sound to be.
Technically, the means by which a sound is converted into a format which a computer can understand relies on a device called an Analogue-to-Digital Convertor or ADC - you can think of it as a Sound-to-Number convertor. The ADC devices generally used to do this work are categorised as either 8, 12 or 16-bit devices. This means that they can convert any number within a given range. Here, 8-bit gives a range from 0 to 256, 12-bit from 0 to 4,096 while 16-bit gives us from 0 to 65,536. By using an 8-bit device, you would have a scale for measuring the level (amplitude) of a sound which would divide it into 256 separate points. So, a sound would start at zero and move up and down in level until it died away (see Figure 1). When you use a 12-bit device, you increase the number of points to 4,096. This means that the different levels the sound reaches can be more accurately recorded (see Figure 2). If you used a 16-bit ADC the accuracy would be phenomenal, with 65,536 different points at which you could measure the level of the sound.
Having established that the higher the accuracy of sample you take, the more faithful a reproduction of the original it will be, we must now consider how often you need to take a sample of the sound.
Remember the film worked at 24 frames per second. If you sampled a sound (rather than an image) 24 times per second, it would result in a tone of 24 Hertz (cycles per second), ie. a very low bass note.
What is happening here is that the frequency (or rate) at which you are trying to sample the sound is interfering with the sound itself. Logic tells you that you can't use a sampling frequency that lies within the human hearing range (generally taken to be 20Hz to 20kHz) because you will obviously always hear the sampling frequency. The answer is to increase the frequency to a point way above your hearing range; for example, compact disc players operate at 44.1 kHz.
Now we must introduce a concept called Nyquist's Theorem. This states that to achieve a clean, distortion-free sound, you must sample at a rate at least twice the maximum frequency you want to record. In other words, if the highest frequency is 20kHz, then you must sample at 40kHz. In general, circuit designers have agreed on 2.2 times as being a better rate due to the filtering of the sound which has to take place just above the highest audible frequency - hence the 44.1 kHz sampling rate found on CD. Figure 3 shows why a 2.2 times rate is better than a 2 times rate. Note how the presence of the filter, which is required to stop any of the sampling frequency (called 'aliasing noise') coming down into the audible range, actually interferes with the upper frequencies of the audio range. However, when the filter is set higher as on the 2.2 times diagram, the effect is reduced. If the sampling rate were a lot higher (eg. 100kHz, as on a Synclavier), you wouldn't need filters at all. It's generally felt that filters compromise the sound quality to a lesser or greater extent depending on their design.
What we have learned so far, then, is that a sound is more realistic when sampled at a fast sampling frequency and stored as a very accurate number.
However, regardless of how well a sound is sampled, once the sound has been captured and stored in the computer's memory you can perform some amazing tricks with it. The sound is stored as a series of numbers and you can play them back in any order. Therefore, you could start half-way though a sound and play only the last part, or only play the middle bit, or loop the beginning. (Remember Paul Hardcastle's 'N-n-n nineteen!'?) It is this scope for sample manipulation, as it is called, which makes one manufacturer's sampler creatively superior to another, regardless of the actual sound quality.
Now that we have covered the basic theory of sampling, let us look at some other important aspects of sampling technology.
In any appraisal of 'audio recording equipment' in a more traditional hi-fi sense, the importance of dynamic range and signal-to-noise ratio (S/N) figures should never be overlooked. However, although we are not comparing a group of tape recorders here, when appraising samplers we are looking at a group of audio recording devices and it is interesting to observe how so very few of the sampler manufacturers actually state these two important figures! Their specification sheets are generally devoid of hard facts about the real audio quality of their instruments. This would never be acceptable in the field of serious hi-fi, where the buying public are more aware of the usefulness of such figures. But the sampler manufacturers seem to have carefully side-stepped the issue and kept the emphasis firmly in the areas where they have always marketed their equipment, ie. the number of facilities, the scope of manipulation, etc.
As a sort of passing recognition to the fact that the sound source is derived from sampling rather than oscillator tone generation, they often slip in a throwaway resolution figure like '12-bit' or the CD quality buzzword - '16-bit'! Some go a little further and say it's '8-bit companding' or something like that. The truth is that bit resolution figures should not be taken as the only criterion by which you judge the sound quality. The general rule of thumb regarding the 'more bits the better' philosophy is, to some extent, a working guideline, but the whole issue of 'sound quality' cannot be measured by that factor alone.
The simple theory related to 'bit resolution' states that you can determine the maximum possible dynamic range of an analogue-to-digital convertor by multiplying the number of bits used by six. Therefore, 8-bit = 48dB, 12-bit = 72dB and 16-bit = 96dB. But is 16-bit the best?
Take the following scenario, for example. A 16-bit sampler would not sound very good if the input signal amplifier circuitry was very noisy and didn't provide enough headroom to take advantage of the theoretical 96dB dynamic range on offer. In these circumstances you could have, for example, 16dB of beautifully digitised noise constantly sitting at the bottom of your dynamic range (the 'noise floor'). This would give you an effective usable S/N ratio of only 80dB. If you were then also faced with an audio circuit which only achieves a 72dB dynamic range due to the poor design and the inferior quality components used - bang goes another precious 8dB! Finally, to make matters even worse, the sounds you are sampling will rarely exist at a constant energy level. This means that certain harmonics present in the sound will be stronger than others. Therefore, the average level of the sound you are sampling will obviously be lower. When you come to set the input record level on your sampler, you need to take this into account and leave yourself enough spare capacity (or 'headroom') to cope with any sudden harmonic transients. Now, if we allow (say) 10dB of safety margin, then our usable dynamic range has dropped to 62dB, with a further 16dB of noise sitting under that! Things are now not looking or sounding so good, are they? And the man in the shop told you it was a 16-bit sampler - "just like compact disc, mate!"
If you improved the design quality of the instrument's audio input circuitry and lost 10dB of the noise at the bottom, you would instantly restore the S/N ratio to 90dB. Again, deduct 10dB of recording headroom and you're left with a very respectable 80dB range compared to the previous 62dB. Remember, both instruments may be 16-bit machines but the difference in sound quality would be remarkable. This is why the number of bits is not always a good guide to sound quality.
Now consider another situation. A company using a 12-bit sampling system (which in theory can only offer a maximum 72dB dynamic range) does some careful circuit design work. They decide to use a 12-bit 'compander' system. This means that the electronics which process the incoming sound work on a similar principle to those of the famous dbx noise reduction system. In other words, they COMpress the sound at the input and then use an exPANDER at the output. This approach usually applies a two-to-one (2:1) compression ratio to the incoming signal and turns a sound, which in reality has a dynamic range of 96dB, into one which in digital terms exists as half of that, ie. 48dB. Now we can fit 48dB quite happily into our 12-bit theoretical 72dB range with no problems and plenty to spare. When the digital data comes out the other end and is converted back into audio, the compander circuit expands the range by applying a 1:2 doubling factor and so restores the 48dB back to 96dB. If such an electronic circuit used high quality components, then it could out-perform a poorly designed 16-bit system.
So, there we have the dilemma. A theory which appears to state the more bits the better, and a practical example of how less bits can sound as good if not better.
Sound quality is also affected by the inclusion of low pass filters at the output stage of the sampler. The method by which sound is sampled means that filters usually have to be employed to cut out unwanted 'aliasing' noise. The Synclavier and Fairlight Series III don't, however, suffer from detectable aliasing noise as they both sample at a very high rate of 100kHz. However, to you and I, filters are a fact of life as they are usually included in all of the more affordable sampling instruments.
If you compared the specification sheets of two different manufacturers' machines and found similar figures - eg. 12-bit resolution at 40kHz sampling rate - why should the quality of their samples sound different when you listen to them?
Well, it all comes down to two basic reasons: (1) the type of filter design used in the system; and (2) the quality of the components used in the audio circuit, as previously mentioned.
Unfortunately, we are all none the wiser when it comes to assessing the audio quality of a sampler. All I have done is to make you aware that the 'spec wars' game isn't always to be won on the bit-number front. As with anything you buy, quality costs money. The Synclavier/Fairlight approach to sampling is one of 'sonic transparency' and offers the very best. Everybody else has to make compromises in order to produce a realistically priced mass market instrument. In the end, your purchase decision will probably be based on two factors: the cost - what you can afford will invariably be fixed by the size of your bank account; the sound quality-your ears are still the best guide in the world!
RECOMMENDED READING: The following two books contain a wealth of information about the theory of sound sampling.
'Musical Applications of Microprocessors'. Author Hal Chamberlin, published by Hayden. ISBN 0-8104-5753-9.
'Introduction to Computer Music'. Author Wayne Bateman, published by Wiley. ISBN 0-471-05266-3.
Feature by Paul Gilby
Previous article in this issue:
Next article in this issue:
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!