Secrets Of Expressive Sequencing, Part 1

PART 1: GUITARS AND DRUMS. Sequencers are great for mechanical grooves — but that's only half the story. As Craig Anderton explains, MIDI sequencers offer tremendous expressive capabilities to those who take advantage of controller editing, logical edit, randomisation, and other common features. This is the first article in a two‑part series.

In this first instalment of our quest for expressive sequences, we'll take a look at two instrument categories, drums and guitars, and cover the sequencing techniques that can make these instruments more expressive. If you need to emulate acoustic instruments in your music, these techniques may well tip the scales into making the listener wonder "is it live, or is it synthesized?"

But First, The Sounds

Sequencing doesn't exist in a vacuum. To create truly great drum sequences, you need equally great drum sounds. Samplers are ideal for playing back drums because the processing options can add more realism. Modulate the filter cutoff with velocity so that hard hits sound somewhat brighter than soft hits; this improves the sense of dynamics.

Another trick is to set a sample's start point several milliseconds after the start of a drum, and use velocity to push the start point closer to the beginning of the sample. This means that hard hits give a louder, more forceful attack; the effect can be much more convincing than velocity‑switching between different samples (although of course, this has its uses as well). While a discussion of drum sound programming isn't appropriate here, just remember that the better your raw materials, the easier it will be to make great‑sounding drum sequences.

Synthesized sounds need to be programmed so that parameters crucial to expressiveness (filter cutoff, envelope attack, vibrato, and so on) are accessible over MIDI, or via physical controllers (mod wheel or pedal, for example). You may find you don't have enough hands and feet to play notes, vary volume, change filtering, and so on; fortunately sequencers let you add these expressive elements after the fact, during the editing process.

Fascinating Rhythms

Drums serve as little more than a glorified metronome in many sequenced tunes, but if you want your parts to swing instead of plod, all it takes is a little extra editing. There are two main qualities that sequenced drum parts often lack compared to acoustic parts: dynamics that let a part 'breathe', and variations that add realism and interest.

• Randomisation Alone Is Not The Answer: Although randomisation can add variations to velocity and timing, human drummers add variations in a non‑random way. In other words, although every hit of a hi‑hat may be slightly different, some of these differences — such as accenting the first beat of a measure — are the result of conscious (or unconscious but nonetheless musical) decisions.

One way to simulate this with a sequencer is to combine randomisation with regional edits. A regional edit is an editing operation that is applied only to notes that fall within a certain region of beats or measures; a typical regional edit might be: "Increase velocity smoothly from 64 to 127 for the notes that fall between bar 4, beat 1, and bar 7, beat 1."

Additionally, change filters (also called logical edit filters) allow edits to be made only to notes that meet certain criteria, such as falling within a particular number of clocks of a specific beat, being within a certain pitch or velocity range, and so on. Adding randomisation creates a randomised overlay within a planned structure; let's look at some examples of drum track edits.

Consider a 16th‑note hi‑hat part played at constant velocity. One improvement would be to use a logical edit operation to increase the velocity of hits that occur on particular beats in a measure, then randomise a bit to add variety. Figure 1A shows how these operations affect a part that originally had all hi‑hat hits at a velocity of 64. The right‑hand column shows the velocity values after the change filter edit; notes that fall on the beat now have a velocity of 84.

Figure 1B randomises the velocities, as well as slightly randomising the start times. Some notes start up to two clocks late or early, whereas some fall right on the beat.

As an example of combining randomisation with a regional change, consider a one‑measure‑long 32nd‑note snare drum roll. Again, playing all hits at a constant velocity sounds really dumb, so add a crescendo by defining the region of the roll and increasing the velocities smoothly from 40 to 125.

However, no drummer can produce a roll that crescends perfectly, with each hit ever‑so‑slightly louder than the last one. Again, randomisation comes to the rescue; varying velocity values within a ±4 range seems to work well as a starting point. Although there will be a general crescendo, the variations ensure that the part won't sound so mechanical. Of course, touch‑up editing on individual notes after these operations can improve a part even further.

• Event Time Editing: Some types of drum patterns, especially those with a particularly idiosyncratic 'feel', benefit from not being quantised. But if your chops aren't accurate enough to give you the feel you want, you can quantise a part and then shape it later, a few notes at a time. Tom fills in ballads, for example, sometimes sound good if they're a few clocks late, as if the drummer is sitting down on the beat rather than pushing it. A crash cymbal hit, on the other hand, might sound best a few clocks ahead of the beat. If you have several percussion sounds hitting on the same beat in the pattern, try pushing the various parts ahead or behind by a small amount; you'll be amazed at how strongly this can affect the feel.
• Timbral Variations: No two consecutive drum hits have the same timbre. As mentioned earlier, sampled drum sounds make it easy to introduce timbral variations that are tied to velocity. But what if you're using drum machine sounds that make the same noises every time you trigger them?

The answer is to create two (or more) slightly different drum sounds. For example, with an Alesis D4, copy the snare sound to a different MIDI note, then slightly detune the copy. There will be a subtle, but noticeable, timbral difference between the two snare sounds. Shift every other note or so of your snare part to trigger the second snare, and you'll end up with a much more lively part.

If you're using sampled drum sounds, try routing a small amount of velocity modulation to pitch, so that high‑velocity hits will have a slightly higher pitch. This gives the feeling of the drum skin being stretched tighter and therefore playing slightly sharp. I usually set the pitch modulation amount so that the change is not really audible except when compared to the non‑bent sound. If you're trying for maximum realism, it's usually better to err on the side of too much subtlety than to end up with an obvious pitch‑bend effect that makes the drum sound less, not more, realistic.

• Signal Processing: Signal processors whose parameters can be controlled via MIDI continuous controllers are another useful tool. Increase the reverb decay time on a crucial snare hit or bring up the EQ in a particular part to accent the toms instead of just increasing their levels. Although it's tempting to enter these changes by manipulating something like a mod wheel, this can eat up sequencer memory and possibly affect the timing of other nearby notes. In most cases you can create any required changes with 'snapshots': sequence one controller event to make the change, then sequence a second event to return the controller to normal. This technique uses very little memory, but more importantly, keeps the MIDI data stream flowing freely by not cluttering it with data.
• Multitimbral Keyboard Consideration: Multitimbral keyboards often include drum sounds along with other synthesized or sampled instruments, making it unnecessary to use a separate drum sound generator. However, these machines do not have unlimited polyphony, and it's a drag to hear notes being cut off when you run out of voices.

I usually find that when programming drums from a keyboard, the note trigger events last longer than needed because of the time my fingers rest on the keys. Many drum sounds will have already played through before the MIDI note has ended, which means that a synth voice may be sustaining for a sound that is no longer playing, possibly reducing polyphony.

On some synths, a voice will become available as soon as its sample finishes playing, so this isn't always a problem. But if you do run short of voices, here's what to do: after recording the drum notes in a sequence, use a 'change duration' command to change all drum notes to the shortest duration required to trigger the sound (a 32nd note, for example). This ensures that no voices will be wasted on sounds that aren't sounding.

If the drum sounds are processed with the same kinds of envelopes as other sounds in the synthesizer, you may need to increase the release times of the envelopes in the drum kit, or set the envelopes to trigger mode, so as not to get short, choppy notes. (Envelope trigger mode may be called something else on your synth — cycle, one‑shot, or non‑sustain, for example. The idea is that the envelope should cycle through all of its stages regardless of when the note‑off is received.)

Stringing Along With Guitars

The easiest way to sequence convincing guitar parts is to find a guitarist who plays MIDI guitar, but that's not always possible. Fortunately, there are several guitar‑specific idiomatic 'cues' that imply to the listener that a guitar is being played. If you add those cues to your sequences, your guitar parts will sound much more guitar‑like.

• I Got Rhythm (Guitar): Most synthesists can fake decent lead guitar parts, but rhythm guitar is another matter entirely. Guitars give wide‑open voicings and the same note often appears in several octaves.

There are three very popular barre chord fingerings (basically first position E, D, and B). In simple rock tunes, the rhythm guitarist will play one basic voicing (usually the first position E voicing) and slide it up and down the neck as different chords are required. Using the same voicing as these barre chords will help create realistic guitar sequences.

Figure 2 shows the keyboard notes that correspond to the three main major barre chord voicings. If the lowest string is not the root, it will often not get played. Also consider that guitarists will often play block chords high up on the neck, and let low open strings ring. These open strings are E and A below middle C, and D above middle C. If these are in your sequence, increase the note length to simulate the increased sustain.

• Strumology: Quantising chords is taboo; when using a pick, it's physically impossible to play all six strings at the same time. What's more, most guitarists will exploit the time differential between the striking of the first and last strings of the strum in a purposeful, musical way.

After examining several sequences containing MIDI guitar chords, the strum timing of chord in Figure 3 seemed most representative for comparatively slow strums. This was recorded at 240ppqn at 120bpm, so the first three notes fall approximately 94, 80, and 48 ms before the beat. The fourth note of the chord lands right on the beat, and the last note of the chord comes about 32ms after the beat. This represents a total spread of around 120ms between the first and last notes of the chord. So, there's an average 24ms difference between each note.

Interestingly, it seems that my playing style tends to 'push' the beat on up‑tempo numbers, as evidenced by more notes falling before the beat than after. This implies that strum timing can be used for 'feel factor' effects that push or lag the overall beat.

• Strumming For The Truly Lazy: Step‑time entry can work well for strums. Set the step‑time duration to a 64th note (again, I'm referencing to 120bpm) and you'll get about 30ms between notes. If that's too much, tighten up the strum by quantising the notes to the nearest beat at about 20% quantisation strength. Randomise the note start times within a 2‑ to 5‑clock window, adjust durations as necessary, and you're set.

If you're really lazy, your keyboard can layer sounds, and the layers can be delayed (the Ensoniq EPS 16 Plus and Peavey DPM 3 SE can do this, for example), you could stack the notes necessary for a particular chord voicing on one key for one‑finger power chord strums. Set the delay for about 20‑30ms between each layered note; to avoid too metronomic a strum, modulate the delay time a little bit if possible.

• A Case Of The Bends: The way you bend notes can make or break a guitar sequence. Don't use the mod wheel, but wiggle the pitch bend wheel instead (remember, guitars don't have mod wheels for vibrato!). Figure 4 shows an up/down bend on MIDI guitar followed by finger vibrato; note the considerable variation between each cycle of vibrato. A triangle wave LFO just doesn't do this. Also note that a guitar string can only bend up, but an LFO modulates pitch up and down around a centre line. The one exception to this involves using a whammy bar, which can bend notes both up and down. Whammy bar vibrato looks like Figure 5.

Generally you'll use whammy bar‑style vibrato on chords, and bend‑up‑only bending on single note leads, but this is by no means a rule. The easiest solution is to have a MIDI guitarist record some pitch bend data into your sequencer and paste it into your own sequences as necessary!

Next month we'll take a look at more expressive sequencing of wind and bass parts.