Basics: Sample Playback Synths

By Jim Aikin

From the very beginning, the synthesizer was an oddity among musical instruments: It had no sound to call its own. The sound had to be specified by the musician, or by a specialist (a synthesizer programmer) hired to do the job. Players quickly discovered ways to make the early analog synths sound vaguely like a string orchestra, a Clavinet, or a marimba, among other things.

The confusion was compounded a few years later, when sample playback instruments came into the picture.

A sample playback synth makes sound by playing back digitally recorded waveforms. Choose a trumpet waveform, and the synth sounds exactly like a trumpet, because what you're hearing is a digital recording of a single note on a trumpet. Choose a snare drum wave, and it sounds exactly like a snare drum, again because you're hearing a digital recording of the real thing. A typical synth includes dozens or hundreds of such waveforms; most of them are digital recordings of actual instruments — trumpets, drums, bassoons, balalaikas, Minimoog, you name it. Others may be computer-generated.

There's a lot more to getting a realistic trumpet sound out of a sample playback synth than just choosing the right waveform and then hitting the keys. You may need to adjust the envelope and filter settings, and you'll certainly want to work out playing techniques that will approximate the behavior of the instrument you're trying to imitate. But that's a whole other can of worms. For now, let's talk strictly about sample playback. What kinds of features will you see in a synth that uses this technology, and what do they mean musically?

RAM vs. ROM

Most of what we'll say in this article about sample playback synths also applies to samplers. The main difference between a sampler and a sample playback synth is that a sampler allows you to load whatever waveforms you like into RAM (a type of computer memory). In a synth, the waveforms will be gathered together at the factory and burned into ROM (another type of memory). The difference between ROM and RAM is that ROM is a permanent form of data storage, while the contents of RAM can be edited in various ways, or replaced entirely with something else.

Why would you want ROM, then? Why don't all instruments have RAM? Good question. In the early days of sample playback synthesis, ROM was cheaper. Also, loading sounds from disk was time-consuming, because most samplers were equipped only with floppy drives. Today that's no longer true, and some synths offer both ROM and RAM for waveforms. But it still takes time to load the RAM — and whatever is in RAM disappears whenever the instrument is switched off. The contents of ROM will still be there the next time you power up your synth. Also, the libraries of ROM waveforms developed by synthesizer manufacturers are optimized to give you a broad and useful sound palette to work from. Some of the time-consuming editing tasks are taken care of for you. With ROM-based waveforms, you get ease of use; with RAM-based waveforms, you get the ultimate in flexibility.

MULTISAMPLING

Switch on your sample playback synth, choose a garden-variety trumpet patch, bypass the built-in effects if your instrument allows it, and play a simple scale up and down several octaves of the keyboard. If you play slowly and listen closely, you'll hear something interesting. Each group of adjacent keys — from three keys to six or eight — will sound exactly the same, except for the pitch of the note. Then suddenly, when you play the next key, you'll hear a slightly different sound, which will be repeated across the next group of keys. One set of keys may have a slightly different amount of "spit" in the trumpet attack, for example.

What's going on here?

In order to conserve memory, a single digital recording of a trumpet note — a single sample, in other words — is being triggered across a range of keys. Some technological wizardry is being used to play the sample at a different pitch, but it's still the same sample. As you move to a higher or lower range, you'll reach a point where suddenly you're triggering a different sample. It's still a trumpet sample, and it was probably created by the same trumpet player, using the same trumpet and the same microphone. But the articulation is slightly different, because a real trumpeter never plays two notes exactly alike. The fact that notes sound identical to one another is a weakness of sample playback synthesis.

The layout of trumpet samples across the keyboard is called a multisample. In your synth's editing pages, the trumpet multisample may be referred to as a single "waveform" with a name like "Trumpet 1," but actually it's a matched set of waveforms. If you listen to the various acoustic instrument multisamples in your synth, you'll discover that some contain only two or three separate samples in different regions of the keyboard, while others may contain more than a dozen closely matched samples.

If you stop to think about it, this is a fairly odd way to design a musical instrument. Usually, one wants a uniform sound throughout the instrument's range. So why not just use one sample for the trumpet, and transpose it up or down to match the pitch of the key being played on the keyboard?

Multisampling is needed because most acoustic instrument waveforms don't sound good when transposed up or down by too many half-steps. Transpose one up too far, and it starts to sound thin and tweezy (the chipmunk effect). Transpose it down too far, and it starts to sound dull and floppy, as if the instrument were built for an elephant and made out of old rubber. Some instrument sounds can be transposed further than others without ill effects. That's why different multisamples in your synth have the samples laid out across the keyboard in different ways.

A few of your multisamples may have different samples played at different velocities. This is a fairly primitive way (in my opinion) to try to coax a sample playback synth into having more realistic performance characteristics. It's called velocity cross-switching. With an electric piano multisample, for instance, you may find that soft keystrokes trigger a sample in which the electric piano itself was played lightly, while a hard keystroke triggers a sample of the piano playing a loud note. If the samples are well matched, this type of multisample can be playable, but if you play a chord at a medium velocity, you're quite likely to hear one of the notes in the chord stick out because it's triggering the high-velocity sample while the keys above and below it trigger low-velocity samples. This type of keyboard response doesn't seem to bother some people, but I don't care for it.

Ideally, we'd like to have a sample playback synth in which each key triggered its own sample — or even better, eight or ten samples at various velocities. That way, no transposition would be needed, and each note would sound its best no matter how hard or soft it was played. The main reason why this isn't done is cost. Memory is expensive, as is the process of preparing the samples before burning them into the ROM, so it makes more sense to sample six trumpet notes than 36, or as many as are needed to cover the entire range of the instrument. With a grand piano multisample, we'd need 88 separate samples just for one velocity layer, which would take up a lot of memory. By spreading each sample across a range of keys and avoiding velocity cross-switched samples, we can fit a lot more variety of multisampled waveforms into the same amount of ROM, thus increasing the versatility of the synth without adding to the cost.

This article presented courtesy of Keyboard Magazine.