Part Two of Newslink's Sampling Course by David
Part 1, in the last issue of Newslink, examined
the basics of sampling - a means of digitally recording and playing back
sound. The technique, rather than recording complete sounds, stores a selection
of samples of a waveform; during playback the missing portions of the sound
are recreated by
from this stored information; and the accuracy of the results largely depends
on the quality (as defined through sample rate and bit resolution) of the
|Fig.1a TRUNCATING SAMPLES
TO SAVE MEMORY
The blank space at the beginning of the sample, where recording started before the sound, will cause a delay on playback; the random dots at the end as the sound tails off will be heard as noise; both waste memory Moving the Start and End points to the new positions indicated will give a cleaner sample and will save memory.
While there are undoubted technological limitations
on quality, processing power is governed more by economics - the more detailed
the waveform analysis, the greater the amount of information to be stored.
Computer memory costs money.
Naturally, in common with all things computer-driven,
sampling technology is becoming cheaper and substantially more powerful.
£5,000 invested today in, say, the Roland S- 770, buys you more than
three minutes of 44.1kHz (CD quality) sampling with 16-bit resolution. However,
the techniques used in sampling were developed 10 years and more ago. Even
five years ago, when the state of the art was the CMI Fairlight Series III,
30 secs of 16kHz frequency sampling would cost you upwards of £48,000
and resolution had only just been upgraded to 16 from the original noisy
With such a paucity of sampling time, early
samplers had to make efficient use of memory. One way was to eliminate all
superfluous data obtained during recording, such as unwanted portions of
the sample itself or blank spaces caused by the recording time exceeding
the sample length. Fig. 1a shows the truncation of sample data using
a digital 'razor blade', in the form of movable start and end points. This
editing was fundamental to the development of the sampler for instrumental rather than purely recording uses.
Unless creative sampling was to be limited to
occasional percussion or effect applications, the musician would need control
over the duration and pitch of the sampled data. Possibilities like the digital
equivalent of a Mellotron, containing inordinately long samples of every
required pitch, were ruled out by cost and memory requirements. The solution
had to allow the conversion of short, fixed-pitch samples to variable lengths
The clue to prolonging a sample lay in the natural
characteristics of instrument wave- shapes. After a brief 'attack' portion
many musical sounds settle to a stable repeating (sustain) waveform before
decaying into silence. The only information required in the sampler's memory,
therefore, would be a record of the attack and decay portions plus a segment
for repetition when the sound was to be sustained.
Choosing the ideal loop points leads to a smooth and natural sound on playback. A difference in value between Start and End points produces a click (known as a glitch. Where no natural looping point can be found, it becomes necessary to adjust the sample data.
The technique of isolating a 'sustain' waveform
for repetition became known as Looping. Two new references, Loop Start and
Loop End, were needed; and the selection of these points would determine
the smoothness of the result (Fig. 1b). In practice, all but a few
samplers only used one Loop point. Looping would take place between the Loop
point and the End point, any key off decay being provided by envelope shaping.
Looping combines two formerly unconnected sections
of a waveform; for optimum results the two points must be at the same part
of the wave cycle, and must correspond in value. Early samplers had only
manual looping. The operator would scan a pictorial representation of the
waveform, searching for likely looping points (the best chance of getting
a smooth join usually being at the peak or trough of a wave). The result
of untidy looping would be a click or 'glitch' on playback.
Modern samplers contain various options to help
looping, such as peak search on the S-series. If manual looping fails to
achieve a smooth join, loop smoothing can be used to rewrite loop point values
and crossfade between them.
The array of playback options - including the unlooped One Shot and Reverse modes, Forward Looping, Reverse Looping and even Alternate Looping - combined with the creative possibilities afforded by truncating and looping waveforms meant that the sampler could be used as a means of sound synthesis as well as a playback machine. Like all instruments, though, it would be of little use without control over pitch.
The key to replaying samples at different pitches
was Sample Rate, or Frequency. Pitch itself is a measure of frequency (see
Basic Synthesis 1 - Newslink Summer 89). Adjusting the playback frequency
would therefore alter the pitch. Two methods were devised: Variable Rate
and Fixed Rate Sampling (Fig. 2).
Variable Rate sampling is the simpler option, in that it does not require the introduction of extra data. It does, however, mean that samples played at lower pitches have fewer reference points. This leads to increased quantization noise (because of inferior sample rate) on playback. Much of the noise will be within the audible frequency spectrum. Filtering the output signal may reduce this, but only at the expense of fidelity.
Variable Rate versus Fixed Rate: Variable rate samplers replay at different pitches by varying the playback frequency. If the pitch Cf (Middle C) were sampled at 44.1kHz then playing back at 22.05kHz (half the frequency) would cause the pitch to drop one octave 10 C3. However, since fidelity is directly proportional to frequency, there would be a corresponding loss of quality: Fixed rate samplers also alter the frequency to adjust the pitch. Extra processing is used to 'create' sample data interpolated from existing values; these are slotted in to keep the sample rate constant and thus maintain sample quality.
| Fig 3
Linear versus Differential Fixed rate samplers interpolate data by drawing a straight line between two known values. This Linear method does not take account of the majority of curved and altering waveshapes. Differential Interpolation (the system Roland uses) analyses several points either side of the 'new' point; the resultant curve sounds much smoother and more faithful to the original.
The more sophisticated alternative, as employed
on Roland samplers, is Fixed Rate sampling. This maintains a high sample
rate, and thus fidelity, through the introduction of assumed (or interpolated)
data between known values. The danger with this method is that any error
in interpolation will start to change the character of the sample. Two methods
(Linear Interpolation and Differential Interpolation) have been employed.
Fig. 3 explains the way Differential Interpolation (Dl) produces more
With the incorporation of editing facilities
plus the elements of pitch and time, the sampler could truly be considered
a musical instrument. Further refinements and extra functions have since
been added, and the availability at a lower cost of processing power and
memory has encouraged users to find new applications for sampling. Before
considering these, the next article will examine some of the early uses of
sampling technology within creative music.
Basic Synthesis 1 | Basic Synthesis 2 | Advanced Synthesis 1 | Advanced Synthesis 2| Sampling 1 | Sampling 2 | A History of Sampling