Sampling with the Kurzweil 250. Will it knock out the Fairlight?
James Betteridge examines the evolution of sound sampling to decide whether the new Kurzweil 250 is a viable alternative to the Fairlight and Synclavier.
Over the last few years sampling synths have increasingly become an accepted part of the professional musician's life. The concept of digitally quantizing a sound and storing it in a solid state memory for use as a musical source is now quite ordinary.
Certain of the original problems associated with this technique still remain, however, as thorns in the side of those same musicians who have to make musical sense from the available facilities, and also for the designers and manufacturers striving to get one jump ahead of the competition.
Memories get larger, sequencer/editing software gets more powerful and improvements are generally made, but the problems associated with the accurate replication of a complex, time-related wave forms, such as the sound of an acoustic piano, have largely stayed with us.
Although it's steadily getting cheaper, digital memory is still very expensive and takes a considerable amount of power.
The use of digital memory is in some ways analogous to the use of magnetic tape. Capturing a transient sound such as a snare drum requires only a very short recording and hence a small amount of memory. But, what is to be done when attempting to record the sound of an instrument such as an organ, which is capable of infinite sustain? Clearly, the cost of providing an infinite amount of memory would severely impair the product's market viability.
The technique employed to get around this problem is also commonly found in all digital pitch changers (harmonisers) and is known as 'looping'. Basically, this means that having read through the stored information once, the 'read pointer' loops back to nearer the front of the memory and re-reads the same information again. This looping cycle is repeated for the duration of the note being sounded.
In the case of 'harmoniser' type devices, the pitch change effect is created by writing the information into memory at a standard speed and then, more or less in real-time information is being read out more quickly than it's being written in: you're going to run out of information.
Consequently, the looping technique is employed. Unfortunately, because of an almost inevitable slight mis-match between the end and the beginning of a sample, each time the pointer loops back, there tends to be a hiccup in the sound. This hiccup is known as a glitch, and is also usually apparent to some degree in sustained notes on sampling instruments.
In the case of a sample-playing keyboard, both situations arise. The cost of recording a separate sample for each of the notes on the keyboard is prohibitive, and so a sample is taken at one or more pitches, and then 'pitch changed' to create the remainder of the notes. As the notes rise in pitch then, the looping rate increases and glitching becomes more rapid. With the very sophisticated software packages available for the likes of the Fairlight and Synclavier systems, the waveforms of a sample can be examined on a VDU, and the loop start and stop points changed. This allows more accurate matching and diminishes glitching problems. Even so, that's only part of the problem, and this looping/pitch change process is far from ideal.
Another weakness lies in the fact that the difference between a high note and a low note on an acoustic instrument, such as a piano, is not contained simply in pitch, it also involves a non-linear change in dynamics and harmonic structure, and this isn't fully taken into account. Consequently, the resultant sound is neither accurate, nor consistent across the keyboard.
As a compromise between unthinkable amounts of memory and low fidelity, better quality instruments utilise a method called 'multi-sampling', where a number of evenly spaced samples are taken across the register, resulting in the need for only a limited amount of pitch changing to 'fill in the gaps'. For example, the 360 Systems keyboard employs 17 samples to produce its high quality acoustic piano sound for which it is well renowned. Interestingly, this company has taken a purist approach to sampling and no looping is used. One drawback here is a limitation in the length of any note's sustain (ie the length of the recording), after which there is a very abrupt cessation.
The attraction of this purist approach is that, while the sample lasts, it is the exact sound of a real piano. However, there is also a second inherent limitation here, in that whilst a real piano is touch sensitive, the fixed sample will provide only the sound of the piano. AS IT WAS PLAYED WHEN RECORDED. This depends on how the musician was playing at the time: hard, and it'll be a bright sound, soft and it'll be mellow. To capture all the dynamic possibilities precisely would appear to require an infinite number of recordings per note, each with a different weight of touch. This wouldn't be overly practical, and so either you're stuck with a static keyboard and no touch response, or standard synthesis methods such as VCA's and VCF's are used to create approximations. Though useful, these have always been very much approximations, and have failed to create a convincing dynamic replication.
This, then, is the dark world of miserable imperfection into which the Kurzweil 250 has been launched. On the surface there is nothing to cause amazement here: it is a MIDI-equipped digital keyboard instrument. It has an 88-note, velocity sensitive keyboard with high quality wooden keys, weighted to closely simulate the feel of a piano. This it does with unusual accuracy, and is one of the most responsive keyboards on the market, assuming you like the touch of a piano. If you can afford this instrument, you can probably also afford a roadie, and hence the substantial 95lb frame (plus 22lb for the power supply) won't put you off.
Mr Kurzweil is apparently a fairly accomplished amateur pianist and enjoys a professional reputation as somewhat of a pioneer in the field of artificial intelligence. The 250 is the result of him applying his knowledge in this field to the aforementioned problems of synthesis.
A large amount of Kurzweil's work with computers has been in the area of pattern recognition, and the system is based on the idea of providing its computer with enough data on an instrument's sound to enable it to form an accurate 'sound model'.
The model is created for the piano, for example, by analysing each note on the keyboard, over the full length of its sustain, and at five different dynamic touch intensities. The piano is actually recorded directly (via a microphone) on to a Sony digital recorder to ensure high initial fidelity, and consistency over a period of analysis.
With reference to this sound model, the 250 can generate the necessary algorithms to accurately recreate the true, time-related dynamic characteristics of the original sound. In the case of the piano, this model apparently includes all the possible combinations of pitch, touch and time factors plus the acoustic interactions relating to the combination of notes being played together — undamped strings will enter into sympathetic resonance to create a unique set of overtones.
The ideal of taking an infinite number of samples is obviously still not possible, and a large degree of extrapolation has been employed. Even so, the results are quite stunning, and include a variety of very convincing string, brass, choral and percussion voicings, including a complete drum kit.
Some form of looping is used but Kurzweil calls it 'timbre extension' and it really is imperceptible (to my ears) over a normal set of speakers.
On top of this there is a built-in, 12-track, real-time sequencer (internally expandable to 7500 notes). It offers a high degree of editability, though it comes as standard with a single stereo output, an optional modification provides 12 separate outputs — one per track — making it ideal for more sophisticated applications. An optional disc storage package provides unlimited off-line storage of sequences.
It is currently available with a single ROM board containing 10 x 256 kilobytes, providing 30 voices. In addition to an almost unlimited key-assign facility, including one voice per key, the voices can, to quite a large extent, be edited by 12, 7-segment envelope generators, plus pitch bend and various time-based effects, although there are no filters as such.
As the Kurzweil library of sounds expands, new voicings will be available on up to three extra internal ROM boards, giving an estimated final total of around 100 internal voices. In addition, ROM cartridges, costing in the region of $850.00 will also be offered containing mostly sound effects.
There are also three software packages promised for the future, all of which will initially be designed for use with the Apple Macintosh computer, which is also based on the 68000 microprocessor. These are to be:
(1) Sound Modelling Program, which will provide 20 seconds of RAM (total) into which your own samples can be recorded. As it hasn't yet been completed, the precise facilities of this software are not known, but it can be safely assumed that home made samples won't match those created at the factory.
(2) Kurzweil Sound Laboratory - basically an additive synthesis package, and even further away from completion than No. 1.
(3) Composition/Notation — this one's also unlikely to hit the shops until late '85 (and that's hopeful). When it does, it threatens to offer a real-time bass and treble clef display of what you're playing, and subsequently a hard-copy print out.
In its present state of development, the 250 can't be compared to the Fairlight, the Synclavier, the PPG or similar, because it has no facility for creating user-sampled sounds. As a 'sample playing' instrument, however, it is out on its own, being unusually effective and simple to use.
It's uniquely gratifying, having written a report of this kind, to be able to offer you the sound of the real thing in cassette form. The Kurzweil 250: the future of digital synthesis? See what you think.