Bob Moog wrote a series of articles about digital sampling, for Keyboard magazine, in the days that digital samplers were a relative new Music technology.
Below you will find the articles (which were completed by Terry Fryer) and we scanned them to include the OCR’d text (below the images).
On Synthesizers digital sampling keyboards,
Part I: a brief history of tape and optical sampling
JUST A FEW YEARS ago, digital sampling instruments were either laboratory curiosities or super-pricey computer-based studio equipment. Today, stratospherically-priced studio instruments like the Fairlight and Synclavier are gaining in popularity, while a new breed of performance-oriented sampling instruments, from the rack-mounted Akai 5-612 to the 88-key Kurzweil, is emerging as the hottest trend in electronic musical instrument design. In this column we’ll trace some of the historic highlights that foreshadowed today’s digital sampling instruments. Next month we’ll have a look at some basic principles of sound sampling, or digitizing. After that, we’ll see what some of the distinguishing features of the currently available sampling instruments are. Optical Recording. Sound sampling is the recording of one or more musical tones in such a way that the tones can be played back instantly on command, usually from a conventionar music keyboard. Today’s sampling instruments are all digital, and as high-tech as anything you’re likely to buy for your own use. But, like so much else in the electronic musical instrument business, sound sampling has its roots deep in the early part of the twentieth century, when virtually all electronic music was experimental. The first instruments that offered playback of recorded waveforms actually used optical recordings on.glass or film. Milton Babbitt, a well-known American composer and one of the founders of the Columbia-Princeton Electronic Music Center, recently told an audience of musicians and doctors about the technology of drawing sound waveforms directly on film for movie score production. This technique was actually used to make the soundtracks for some· Russian movies before World War II. Of course, waveforms on film could be played back immediately only in the sense that they could be precisely synchronized and assembled by conventional cinematic editing techniques. Babbitt’s account is a fascinating glimpse into the multi-faceted world of experimental music. I’m planning to make it the subject of a whole column in the near future. Optical sound recording technology was applied to commercial performance instruments after World War II. During the early ’50s, when I was a teenager just getting into electronic musical instruments, the Baldwin piano company introduced an electronic organ containing several rotating glass disks, on which were recorded, on circular tracks, the waveforms of actual organ pipes. There was a light source on one side of the disk and photoelectric detectors on the other side. When you pressed a key, you moved a shutter that allowed the light to pass through the disk at one of the tracks. This design not only provided rich, complex waveforms, it solved the problem of smooth attack generation. Back then, most electronic organs were not able to produce tones with slow attacks, but the experimental Baldwin design tied the shape of the attack to how fast you pressed the key-not unlike the classical tracker organ. As it turns out, Baldwin’s optical organ never made the big time, primarily because the disks had to be made to incredible accuracy in order for the tones to be free of wow and flutter. It was a mechanical nightmare-big, super-precision spinning disks with dozens upon dozens of flexible linkages going to the keys. The same basic idea surfaced again about twelve years later in the Mattel Optigan. Mattel developed a line of ‘fun ma<;hines’ with short keyboards, tacky plastic spinet-like cabinets, and big flexible transparent disks containing a few dozen waveform tracks. Most of the tracks were of pitched notes, but a few contained rhythms and novelty stuff. The idea was that you could change disks to suit the kind of music you wanted to play-country and western, foxtrots, and so on. It was fun to play the rhythm tracks. You could switch rhythms and drop in riffs with one hand while playing chords and melody with the other. Unfortunately, even though the Optigan was scaled to be a low-grade adult toy, the manufacturing problems proved to be insurmountable, and Mattel chucked the project. David Van Koevering, an enterprising musical instrument salesman, pulled the basic Optigan technology out of the ashes and introduced an instrument called the Orchestron, which he aimed more at the rock-and-roll keyboard player. With an Orchestron, you could play strings and voice choirs, a musical capability that was very salable at the time (and still is!). Once again, however, the realities of manufacturing such a finicky mechanical system conspired with the usual financial and marketing demands that are placed on innovative musical instrument companies, and the Orchestron, too, passed into history. Magnetic Tape. Experiments with taperecorders and electronic music began in earnest after World War 11. One experiment done in Paris was the Phonogen, a true sampling machine with keyboard control and high sound quality. The Phonogen played a regular magnetic tape, either off a reel or spliced into a loop. Instead of having one capstan to drive the tape at a constant speed, however, there were twelve or thirteen capstans in a circle, all going at once at speeds whose ratios were the same as the frequency ratios of the notes in a chromatic scale. Each capstan had its own pressure roller, and each roller was linked to a key on a little one-octave keyboard. When you pressed a key, you pressed the tape against one of the capstans, thereby causing the tape to play back at one of twelve chromatically related pitches. With this system, you had electronic sampling (onto tape) and real-time playback under keyboard control. What you didn’t have were the ability to start a note at the same point on the tape every time, and the ability to play chords. Both of these shortcomings of the Phonogen were dealt with in the Mellotron and the Chamberlain. These instruments used a whole set of short tapes, each of which was activated (that is, pressed against a constant-speed capstan, which pulled it past a playback head) by a key of a short but conventional key_board: In Bob Moog both machines, a tape started from its beginning each time you pressed a key, and rapidly repositioned itself to the beginning as soon as you let go of the key. Thus, you could play the starting transient of a complex sound if you were careful not to repeat a note too rapidly. Chamberlain’s machines were well known and widely used by West Coast musicians during the ’60s and ’70s. The Mellotron, on the other hand, was professiona 11y marketed while rock and roll went through its growth phase, and by the early ’70s had become an indispensable staple of the rock keyboard arsenal_, despite its mechanical complexity. Enter The Digital Era. At their best, analog sampling instruments-the Chamberlain and the Mellotron-produce sound whose quality is marred only by the usual tape noise and flutter. You can play any number of notes and hold individual notes for several seconds, until the tapes run out. Eventually the tapes do wear out or break, so you replace them just as a guitar player periodically replaces his strings. What else could a keyboard player want? Less hassle with finicky mechanical systems, for one thing. But more than that, the electronic keyboard player of the ’70s was becoming aware of how nice it was to have a whole palette of tone colors to choose from. With the Mellotron you had a couple of sounds that you could use on a set of tapes, but changing tapes during a gig was unthinkable. And how about layering orchestral sounds to create those lush textures? Or electronic modifications like vibrato and chorusing? Something more was needed. If you combine the musical concepts of the synthesizers of the late ’70s (rapid program change, electronic sound modifications, keyboard layers and splits) with the intrinsic sonic complexity and richness that the analog sampling machines offered, you come up with the 1 set of capabilities that the current crop of digital sampling instruments offers. And that’s what we’ll start talking about next month.
On Synthesizers Bob Moog sound sampling instruments
Part 2: technical specs & sound quality in digital sampling
LAST MONTH WE TOOK a quick tour through musical techno-history, stopping briefly to examine the roots from which today’s sJmpling instruments have grown. We saw that, technically speaking, there are many ways to ‘skin the cat’ of.sound sampling, from drawing the sound’s waveform on a glass disk to using large semiconductor memories to store digital encodings of complete musical sounds. Today’s instruments that let musicians sample their own sounds are all state-of-the-art microprocessor-based digital devices with a random access memory (RAM) capability that exceeds the storage capacity of your average personal computer, and with provision for some sort of off-line non-volatile (capable of being saved indefinitely) storage, usually on floppy disks. These instruments’ operating software, usually contained in one or more ROMs (readonly · memories), contains a series of sophisticated routines (small,self-contained programs that perform specific operations). Together, these are able to do everything necessary to digitize, process, and play back musical sounds. Those who are familiar with the basic concepts of computer architecture recognize that a digital sampling instrument is not that different from a general-purpose personal computer. In fact, you can buy peripheral-plus-software packages for many personal computers that let you sample sounds from a microphone or tape recorder and perform elementary operations on them. The important differences between these peripherals and dedicated sampling instruments are all in the details of the user interfaces, sampling hardware, and operating software. Well-designed sampling instruments are optimized to produce high sound quality and offer ergonomically appropriate control over the sounds, all at a reasonJble cost-the same optimization that is embodied in all popular musical instruments. Some Basic Sampling Concepts. In digital audio, digitizing is the conversion of an audio waveform into a series of numbers that stand for the height (instantaneous amplitude) of the waveform at closely spaced intervals of time. Each of these numbers is called a sample. It is the value of the waveform at one fleeting instant of time, an extremely thin slice of the waveform. The piece of hardware that does the sampling is called an analog-to-digital (A/D) converter. These days, the A/D converter is a single-chip device whose input is the audio waveform and whose output is a set of off/on signals that represent the binary bits of the sample. There is one other connection to the A/D converter that you should know about. It is often called the strobe input. If you think this has anything to do with strobe lights, you’re right. When the strobe input receives a pulse, it grabs the value of the waveform at that instant and converts it to a sample number. In much the same way, a strobe light ‘freezes’ the motion of an object (such as a dancer). Fig. 1 shows how an engineer would draw an A/D converter. The strobe input pulses thousands of times every second, and its frequency is called the sampling rate.
Another analogy might be helpful in understanding the numbers associated with digitizing. That analogy is with video and film. Like digitized sound, film is made up of a series of still images (samples). l.f we see the images rapidly enough in succession, they appear to move smoothly; if the images don’t change rapidly enough, the movie is jerky and disjointed. In a roughly similar way, sound samples appear to be indistinguishable from the original waveform only if they follow one another in rapid enough succession. Now let’s relate the sampling rate and resolution of digitized sound to those gremlins that plague electronic musicians: noise and distortion. Sampling Rate. Unlike our eyes, which have a long ‘persistence of vision’ that lets our brains fill in between the frames of a movie or TV program, our ears hear virtually every change in a sound that produces frequencies between 20 and 20,000 cycles per second (Hz). When we digitize a sound, we first have to decide how much of the frequency spectrum of that sound we want to preserve. Then we have to be sure that the sampling rate is fast enough to preserve all those frequencies through the sampling process. How does one discover the relationship between sampling rate and frequency response? Fortunately, Bell Labs scientist Claude Shannon studied this relationship forty years ago, when digital transmission of analog data was just in its infancy. By using some pretty tricky math, Shannon showed that the sampling rate must be at least double the bandwidth of the digitized signal in order for all of the information in the signal to be preserved. This makes intuitive sense as well: If a waveform is cycling from high to low 20,000 times every second, we need to take at least one sample every time it is high and another every time it is low, in order to get any sort of picture of its shape. In other words, we need to take twice as many samples as the wave has cycles. When engineers design a digitizer, they use sampling rates somewhat greater than Shannon’s criterion. For instance, a professional digital audio recorder runs at a sampling rate of 48k Hz, even though the bandwidth of the audio is no more than 20kHL. In sampling instruments, the sampling rate may be anything from 5 or 6kHz (to digitize a sound that has very limited bandwidth) to 100kHz (which digitizes well beyond the range of human hearing). However, the sa rn pli ng-rate-to-f requency-bandwidt h ratio of 2.4 or 2.5 to 1 is nearly always used. What does this mean for typical musical material? Well, a single musical note of, say, l0kHz bandwidth and 3 seconds’ duration requires about 70,000 samples to digitize. Storing that one single sample takes more RAM than is in the entire personal computer on which I’m writing this column! As we’ll see in a later column, the sheer amount of RAM in a sampling instrument is one of the most important factors in its cost. Accuracy (Resolution) Of The Sampling Process. The more bits are used to record the value of a sample, the more accurate the sample is. A good analogy here is the representation of a black-and-white photo as a grid of dots, where all the dots must be either black or white (ones or zeroes). Using a few large dots will give a very poor idea of what is in the picture, while using lots of small dots packed cl03e together will give a much better representation. Obviously, translating shades of gray into black and white causes some error. The errors in sound sampling add up to unwanted information: noise (if the errors occur randomly) or distortion (if the errors are related to what the desired signal is doing), A good rule of thumb is that every extra bit of resolution in the sample contributes 6dB to the signal-to-noise ratio. An 8-bit sample thus has a signal-to-noise ratio of about 48dB, while a 16-bit sample has a signal-to-noise ratio of 96dB. In terms that we are familiar with, 48dB is what you get out of a run-of-the-mill cassette recording with no noise reduction, while 96dB is stateof- the-art compact disc quality. Sampling resolution is one of the easiest parameters to talk about when comparing two sampling instruments. Other things being equal, a 12-bit machine will offer better sound quality than an 8-bit machine. But other factors can have a major effect on the quality as well. Some sampling instruments offer only linear response (the noise stays the same as the signal level goes down) while others offer logarithmic response or compression/expansion (the noise level goes down as the signal level goes down). These differences, plus the differences in the instruments’ input filters, which limit the frequency bandwidth of the inrnming signals (and al/ digital sampling instruments must hJve input filters), account for much of the individual sound character of the currently avail,1ble sampling instruments. We’ll discuss these and other small but important details of sampling instruments in next month’s column.
On Synthesizers sound sampling instruments
Part 3: quick listening tests for digital samplers
LAST MONTH, WE delved into the technical side of sound sampling. We saw that sampling in a digital instrument consists of taking a rapid series of snapshots (samples) of a sound waveform, storing the snapshots as digital numbers, and playing them back to reconstruct the original waveform. In order to sample a sound, the samples must be taken at at least twice the rate of the sound’s bandwidth. Noise and distortion in a sampled sound are related to the resolution (number of bits per sample). If each sample is eight bits, then the sample’s noise and distortion can be no better than 48dB below the desired signal; for every extra bit of resolution, the theoretical best signal-to-noise ratio gets better by 6dB. It’s important to understand that the above signal-to-noise figures are theoretical ‘bestcase’ numbers. Real-world sound sampling instruments generally produce more noise and distortion than the best-case numbers would lead you to believe. The reasons for this extra noise have to do with details of an instrument’s hardware. For instance, sampling instruments that are built with only one sound channel generally produce more noise than instruments that are built with a separate channel for each voice. This means that two sampling instruments that both use eight-bit sampling may sound distinctly different from each other, or that an instrument with ten-bit sampling may seem to have higher fidelity than another with twelve-or even sixteen-bit sampling. As a musician who is interested in comparing the sampling sound quality of several instruments, you should know a couple of simple listening tests that you can perform in a few minutes, without a lot of equipment, which rely primarily on your ears. Sampling, and then listening to, a conventional musical sound (such as a synthesizer tone or a cymbal crash) will tell you something about the quality of the sampler. Noise and distortion in the sample will usually cause the output to be dull and muddy. However, it is possible to construct sounds that more clearly show how much and what kind of noise and distortion are being produced. In order to run these tests, you will need some sort of small synthesizer that is capable of producing sine-like as well as buzzy pitched tones, and of imparting slow frequency modulation (a siren effect from a sine or triangle wave LFO) and amplitude contouring (enveloping) to the tone. Of course, the instrument that you use to generate the test tones should not produce audible distortion or noise of its own. A small modular or monophonic analog synthesizer is ideal. Test No. 1. Set up the test synth to produce a single sustained tone, of a frequency around 1 kHz (three octaves above Middle C), and of low harmonic content. If your instrument produces a sine wave, use that; if not, then use any waveform and set the lowpass filter so that the overtones are filtered out, letting only the fundamental frequency through. Your test tone should be pure and flutelike, with no audible noise or distortion. Once you’ve set up your test tone, sample a couple of seconds of the tone with the sampling instrument you are testing, following the manufacturer’s instructions to the letter. Pay particular attention to the input level setting, to insure that your test tone is loud enough to give a good sample but not so loud as to overload the input electronics of the sampler. Now connect the output of the instrument under test to one input of your monitor system, and the output of the test-tone syn th to another input. By switching back and forth, compare the sampled sound to the original. You will clearly hear the difference between the two. Since the test tone has only one frequency component, the difference that you hear cannot be attributed to the frequency response of the instrument under test. The difference has to come from added noise and distortion introduced by the sampling process. Test No. 2. To hear what the noise and distortion components are in some detail, repeat Test No. 1 with one additional element. This time, frequency-modulate the test tone with a slow (one cycle every second or two) sine, triangle, or sawtooth LFO. The pitch of the test tone should go up and down about an octave. When you play back the sampled sound and A/Bit with the original, you’ll hear three types of additional stuff that aren’t in the test tone: additional harmonics that just brighten the tone, ‘whistles’ that don’t follow the pitch of the test tone, and hissing or rumbling that seems not to change as the test tone’s pitch rises and falls. Let’s look at these one at a time. Increased brightness in a tone that is otherwise clean is a sign of harmonic distortion. This is the least bothersome form of dirt, since the distortion is harmonically related to the desired pitch. For nearly all traditional musical tones, a small amount of harmonic distortion is usually not noticeable. However, occasionally you’ll hear a high-pitched buzz or whistle that perfectly tracks the test tone’s pitch. This is called high-order harmonic distortion, and it is objectionable, especially when flutelike and similar tones of low harmonic content are sampled. Whistles that do not track the test tone’s pitch are alias components-interactions between the test tone and the sampling frequency Bob Moog that are not adequately filtered out. Alias tones are perhaps the most objectionable type of garbage that a sampling instrument can produce. Alias tones are heard as distinct pitches when pitched sounds are sampled, or as an overall muddiness when percussive or noisy sounds are sampled. Either way, the effect is musically distracting, unless you happen lo be into exploiting the specific sound modification resources of aliasing. (One person’s ‘sonic excitement’ is another person’s ‘annoying racket.’) Hiss, rumble, and similar unpitched background sounds that appear not to change as the test tone’s pitch rises and falls are the result of either noisy circuits, errors from the limited number of bits per sample, or errors in the timing of when the samples are converted to waveform points. They are generally less objectionable than aliasing, but more objectionable than harmonic distortion. If the loudness is the same for either a strong or a weak sampled sound, then it is certainly more objectionable than would be the case if the background distortion became softer as the desired sampled sound became softer. Test No. 3 sheds light on this aspect of background noise. Test No. 3. Repeat Test No. 2, but (a) speed up the frequency modulation to about two cycles per second, and (b) put an envelope with a fast attack and slow decay to zero (a couple of seconds) on the test tone. Sample the entire sound. If the entire sound doesn’t fit into the memory of the instrument under test, then shorten the test tone’s envelope. Listen to the background noise as the sampled sound drops to zero. The loudness of the noise may go down with the loudness of the test tone. Or the noise may stay at its original volume, and appear to swamp the tail of the test tone’s decay. Even worse, the tail may become increasingly crackly and distorted, then suddenly fall into complete silence or otherwise change character abruptly. The first instance is the most desirable behavior, since it is the least noticeable in sound samples that have long attacks or decays. The second instance is just like noise in most analog devices. It’s certainly not desirable, but it can usually be managed. The third instance arises from a combination of too few bits per sample a”nd overly simple circuitry that inadequately handles low-level portions of sampled sounds. In any sound with slowly varying amplitude, this last behavior is definitely musically distracting. Next month, I’ll give you some more simple listening tests for sampling instruments.
Digital sampling the roots of keyboard sampling
Terry Fryer is a principal in Co/not/Fryer Music, Inc., a commercial music production company in Jhe Chicago area. Their accounts include Budweiser, MacDonald’s, United Airlines, and Doritos. Terry is also president of Ear Works, a 24-track synthesizer studio devoted to sampling keyboards.
YOU’VE SEEN THE ADS. “An acoustic piano in a box.” (Portable-not just a piano with a handle on it.) “A recording studio in a box.” “Any sound you can hear or think of,” in a box. It may all seem confusing at first, but part of the fun is discovering how simple the ‘magic’ actually is. What we’re talking about is the onslaught of the digital sampling keyboard. They range in cost from the price of a three-bedroom house to under two thousand dollars, and while some are bulky enough that you hope a pair of large roadies are lurking nearby, others can easily be tucked under your arm as you run out the door to your next job. Fortunately, they all have a lot of things in common as well as a body of terms that have come into use to deal with sampling and its technology. If you’ve been following Dominic’s Basic Synthesis column [Keyboard, Aug. ’83 to May ’85), many of the concepts and terms you already know can be applied to these instruments. I will be introducing new terms and concepts in a practical context. So whether you’re in the market for a sampling machine, just curious, or even an accomplished sampler, this column will strive to provide answers to your questions about why these machines work the way they do and how to get them to do what you want them to do. (If you’re in a hurry, check out “Digital Sampling Keyboards” in the Dec. ’85 Keyboard.) What Is Sampling? A dictionary definition is “a portion, piece, or segment that is representative of a whole.” In statistics a sample is a set of elements drawn from and analyzed to estimate the characteristics of a population. If we substitute “sound event” for “population,” we have a workable description of what these machines can do. They don’t actually record the totality of any sound. All they do is latch onto a representative sample. A short look into history will help explain how this came about and give you lots of good material to listen to and get ideas from. In 1948, a technician working at the Radio-diffusion Television Francaise in Paris produced several studies that he called musique concrete. His name was Pierre Schaeffer. The idea was fairly simple: Take a microphone and a recorder (he originally used discs), run around the countryside recording anything that sounded interesting, get several boxes of razor blades and splicing tape, and spend several thousand hours editing the sounds together. Many of the techniques developed at the research center are implemented in today’s sampling instruments. Cross-fades, layering of sounds, multi-tracking, splicing, looping, and playing sounds forward and backward at various speeds all have their roots in tape manipulation and can now be d<;>ne in a fraction of the time (and of course without splicing tape). The list below gives some good examples of musique concrete and a couple of sources that describe some of the techniques. As we move through this topic in the months ahead, we’ll show how these moves are done with tape in the analog realm and then with their digital cousins, the sampling instruments. . In the mid-’60s, the technique of direct computer synthesis was introduced. What made this possible was the development of the digitalto- analog convertor (DAC) and the availability of large digital computers. A DAC takes a string of numbers from the computer and turns them into a signal that can be recorded on tape or directly listened to. Now instead of running down to the physics lab and pinching a sine wave oscillator, it was possible to input the formula for a sine wave into the computer, and presto-a sine wave came out the other end. Elaborate formulas have been evolved to produce complex tones and even mimic acoustic instruments. Since then, computers and computer memory have gotten cheaper. By combining this power with the techniques of musique concrece and direct compute-r synthesis, we arrive at sampling as we know it today. Instead of sending the signal from the microphone to a taperecorder, send it through a reverse DAC, or analog-to-digital converter (ADC). Now the numbers that are stored in the computer memory are not the result of a mathematical formula but are directly related to a real-world sound. When this is played back, the DAC produces an event that is supposed to sound like the original. Unfortunately·, ‘a couple of problems arise, and the sound may not be as perfect as you want it to be. Sound is a continuous medium, meaning that a sine wave with a frequency of 1kHz moves without breaks or steps through its motion 1000 times every second. Digital computers, on the other hand, do not operate in a continuous manner; they deal with incremental steps like 0 and 1. This is where the terms “resolution” and “sampling rate” become important. Back to our 1k sine wave-let’s give it a voltage swing of +1 to -1 volts. When we’re in a tape medium, it’s relatively easy to record and reproduce this tone. The tape machine is put into record, the tape rolls, and the record head magnetizes the tape. As the tape is played back, the playback head responds to the magnetic flux stored on tape and gives us back our nice, continuous sine wave. In the digital domain some decisions need to be made. Since these computers think in terms of discrete steps, the first decision is one of resolution-into how many pieces should we divide the voltage swing? Not enough and we’ll get a jagged waveform which does not resemble the original either by sight or sound. Too many, and the cost of the DAC, ADC, computer memory, and loss of processing speed becomes prohibitive. The second involves the sampling rate-how many samples do we need to take each second? The same kinds of problems Terry Fryer occur. Too few samples, and the recorded version doesn’t resemble the original. Too many samples and the memory requirements go out ·of sight. In the months to come, we’ll examine these issues in more depth, bringing to light some well-loved terms such as Nyquist limit, aliasing, quantization error, and the like, and see how manufacturers and musicians have dealt with these liabilities. Next month we’ll cover how resolution and sampling rate affect the sound that is sampled and how to figure out how much memory. is needed to get the optimum sample. While you’re waiting for next month’s column, listen a bit to what some talented people have done with some stone-age equipment. Be encouraged, though-the same techniques they used can be employed on most of the available sampling units in a fraction of the time without slicing up any tape (or your fingers). Remember when listening to these examples that these were all constructed from little pieces of tape, laboriously hand-spliced together. If you’re adventuresome and want to get an idea of how much of a time-saver a digital sampler is, try your hand at some of these techniques.
Recommended Listening: LeCaine, Hugh: Dripsody (1955), Folkways Records (Folkways/Scholastic, 632 Broadway, New York, NY 10012) 33436. Schaeffer, Pierre: Ewde aux chemins de fer (1948), Ducretet-Thomson (address not available), 320 C 100. Stockhausen, Karlheinz: Gesang der }unglinge (1955- 56), Deutsche Grammophon (c/o Polygram, 137 West 55 St., New York, NY 10019) DG138811 (1955- 56)-a combination of electronic and natural sounds. Stockhausen, Karlheinz: Anthems For Electronic And Concrete Sounds (1957), Deutsche Grammophon, DG2707039. Ussachevsky, Vladimir: Sonic Contours (1965); Luening, Otto: fantasy in Space, Low Speed, Desto Records (c/o CMS Recording, 226 Washington St., Mt. Vernon, NY 10553) 6466. Varese, Edgard: Deserts (1954), CR I (170 West 74th St., New York, NY 10023), 268. Xenakis, lannis: Bahar (1962), Concret P-H (1958), Orient-occident (1959-60), Diamorphoses (1957), Nonesuch Records (c/o Electra/Asylum, 665 Fifth Ave., New York, NY 10022), 71246. Recommended Reading: Repertoire International des Musiques Electroacoustiques, compiled by Hugh Davies (M.I.T. Press, Cambridge, MA, 1968). This book contains works created before April 1967. Davies, Hugh,” A Discography Of Electronic Music And Musique Concrete,” Recorded Sound No. 14 (April, 1964). Ellis, Merril, “Musique Concrete at Home,” Music Educators Journal, Vol. 55, No. 3 (November, 1968). Judd, F.C., Electronic Music And Musique Concrete, Neville Spearman, 1961. Schaeffer, Pierre, A la Recherche dune Musique Concrete, (Editions du Seuil, Paris, 1952). Ussachevsky, Vladimir, “Notes On A Piece For Tape Recorder.” In Problems Of Modem Music, edited by Paul Henry Lang (W.W. Norton & Co., New York, 1960).
Digital Sampling sampling resolution bit by bit
FOR THOSE OF YOU who were adventuresome after last month’s column and tried some of the musique concrete experiments suggested in my bibliography, I hope your· fingers have healed. For those who missed last month’s column, we talked about the origin of sampling m~shines in the methods of tape manipulation used in avant-garde classical music in the years immediately following World War II. For you armchair quarterbacks, take my word for it, there are a lot of things in life better than slicing up microscopic bits of tape at three in the morning and then trying to decide if they’re going in the right direction. Last month we introduced the idea that sound is a relatively continuous medium. A 1 KHz sine wave moves smoothly through its cycle with no bumps or ridges. The keyword here is smoothly. An analog recording device, such as an ordinary tapedeck, will capture a wave continuously as it changes amplitude. There are no divisions that we can recognize, any more than a surfer could recognize divisions as he moved up over a wave. A digital sampling keyboard cannot make such a recording. It is a computer; it must record amplitude changes in steps. Computers cannot handle continuous information. Unlike the smooth sine wave we are used to, the digital version is jagged, like a staircase. The first question is, how many steps will it take to represent a continuous musical waveform and make it sound good? The term resolution is used to refer to the number of steps into which a sampling machine divides the amplitude of a waveform. Each step has to be represented by a number the computer can understand. Computers understand binary numbers best. These are composed of zeros and ones. Each computer is limited by the number of these zeros and ones it can handle. A one-bit computer can make two such numbers. Obviously, the more computer power we have, the smoother the curve begins to look and the closer we get to the original sound. Let’s take a look at some of these waveforms broken into digital form. Ex. 1 is a representation of three-bit linear resolution. A three-bit computer can make eight discrete steps out of the wave. Not exactly a surfer’s delight. You can rest assured that it doesn’t sound much like the original either. This problem is called quantization error, and it produces quantization noise. A good description borrowed from photography is “grainy.”This is the digital version of tape hiss. As we move through the examples, however, it becomes apparent that all is not lost. Our imaginary surfer has less stair motion and more wave motion aswe move to four, Ex. 1. 3-bit resolution 4-bit resolution 5-bit resolution r / &-bit resolution five, and six-bit resolution. The final diagram (Ex. 2) in this series is seven-bit resolution, which generates a waveform that looks quite a bit like the original. But things get even better-looking (and better-sounding) with 12-bit and 16-bit resolution (Ex. 3). Instead of the 256 possible choices for representing the waveform’s amplitude that we get with an eight-bit machine, by using a 16-bit machine we have over 65,000 discrete steps. If you want to know what that means in terms of actual sound quality, you can compare the resolution of a good sampling machine to compact disk players, which use either 14 or 16-bit resolution. Terry Fryer Ex. 2. 7-bit reso!ution Why do compact discs and 16-bit samplers sound so good? One reason is that resolution determines the maximum dynamic range of the instrument. In this case, dynamic range means signal-to-noise ratio. But simply put it means that great resolution gives you the widest possible range of sounds you can sample and reproduce. A lot of mathematics can be dredged up to compute the maximum signal-to-noise ratio based upon how many bits of resolution are available, but they boil down to a simple rule of thumb-multiply the number of bits by 6dB. With a little arithmetic you can readily see that an 8-bit machine is capable of a 48dB signal-to-noise ratio, and a digital sampling keyboard with 16-bit resolution is capable of 96dB. In other words, with higher resolution, you’ve got a lot more room to play with before you start hearing noise instead of what you’re trying to record. Now everything I’ve said so far exists in a perfect world. Machines never work the way mathematicians think they do, however. In the real world, strange things begin to happen. For example, all the steps that the digital sampling keyboard uses to represent the amplitude of the waveform may not be the same size. The analog-to-digital converter might stumble now and then going up quirks that come into the process, the more noise you get. So the specifications you see are ideals: They show you what can happen when everything is working exactly right under laboratory conditions. So now you have enough information so that you can look at some spec sheets and feel like you understand something. At least you can talk to the salesperson. Or can you? When you start scanning those spec sheets for popular digital sampling keyboards, you’ll quickly see that all things are not equal. E-mu claims an 8-bit proprietary encoding scheme that is the equivalent of 14-bit linear resolution. Kurzweil lists an 18-bit modified floating-point scheme which gives a 96dB dynamic range. (By using our rule of thumb, that indicates something like 16-bit resolution.) Why all the confusion? All my previous examples are based on linear models; that is, nothing tricky has been done in breaking the smooth wave up into steps. The digital sampler attempts to break it into as many equal steps as it can. But there are several other types of data encodin·g and decoding. Some neat mathematica I tricks can be used to increase the resolution (and therefore the sound quality} • of the system. For example, the use of digitalto- analog converters and analog-to-digital converters that respond in an exponential rather than linear manner, or a process called companding, where the signal is compressed on the way in and expanded on the way out. Another device is to use any of a variety of floating-point arithmetic schemes. For the math masochists reading this, there is an excellent book that goes into great detail on encoding schemes as well as a great many other computer topics, Musical Applications Of Microprocessors, by Hal Chamberlin (Hayden Book Co. Inc., Rochelle Park, NJ 07662] . . Fortunately for the rest of you, you don’t need to know all that. The standard for comparison is always either dynamic range (signal- to-noise ratio in dB} or its equivalent in linear resolution (8-bit linear, 12-bii linear, and so on}. A little mental calculation with our rule of thumb will convert dynamic range to linear resolution and back again. Now you know what resolution is and how it affects dynamic range. You can read a specsheei, make a rough estimation of a digital sampling keyboard’s dynamic range, and get some idea of how good it will sound. But there’s more to sampling than reading a spec sheet. How does all this apply once you start recording into your sampling keyboard? Record at the hottest possible level without distortion. If you have a visual editing system, it’s pretty easy to eyeball the waveform and see if you’re taking advantage of the full dynamic range of the instrument. The reason for this is simple. Take another look at Ex. 2. In the crude 3-bit machine, there are only eight possible steps to represent a waveform. If you recorded something with that machine and used only half the volume that was possible, you’d only get four steps or three to represent the amplitude of the waveform. Now let’s take a more realistic example, an 8-bit machine. By using the rule of thumb, we know that an 8-bit machine has a dynamic range of 48dB. Using only half of the steps means using only half the dynamic range. When you replay it and turn the volume up, you’ll be turning the noise up as well as the signal. Since the object is to record the quietest, cleanest, most realistic samples possible, record them hot.