Even the biggest sample library never seems to have quite enough drum sounds. The solution? Roll your own. Craig Anderton passes on some hot hints and tips on how to get the best possible drum sounds out of your sampler.
Drum sounds and samplers make a super combination, not just because it's possible to sample virtually any drum sound you like, but also because drum sounds can usually live with sampling's main limitations (limited memory, and looping). Apart from cymbals and sounds with lots of ambience, drum sounds don't require a lot of memory and generally don't need to be looped.
However, just getting a drum sound into a sampler isn't enough. Real percussion instruments offer a great deal of expressiveness and dynamics - qualities that need to be simulated with the sampler's synthesising/signal processing capabilities (filters, amplitude envelopes, etc.). And since there's never enough memory, even for drum sounds, it probably wouldn't hurt to have your drum samples take up the least amount of memory possible. We'll address these and other topics in this article.
We'll make a few basic assumptions: that you already have a collection of raw drum samples just waiting to be optimised, and that your sampler can make copies of existing samples. There are two ways to do this: by 'cloning' the data of an original sample, or creating an entirely new sample with wavedata that just happens to be the same as the original sample. The first approach saves memory and is recommended, although if you edit the original sample, you'll edit all the clones as well. Copy processes that make completely different versions of the same sample allow each to be edited individually. Most samplers include both copy options.
The ability to layer samples on the same key is also crucial to many of these tips. Finally, some of the following techniques are much easier to accomplish if you have a visual editing program (eg., Avalon, Sound Designer, Alchemy, SampleVision, Sample Wrench, Genwave, etc.). Ready? Let's start.
It's more fun to play sampled drum sounds if they respond well to dynamics.There are several ways to improve dynamics with the sampler's on-board sound modifiers.
Start by tying the overall drum level to velocity. Check out the different velocity curves your sampler offers; some will give a better 'feel' than others for percussive triggering. However, also making the timbre track velocity can provide an even more realistic sound.
There are two basic ways to do this:
• Lower the filter cutoff somewhat, and assign it to velocity. Low velocity sounds will have more filtering and sound 'darker' while high velocity sounds will be 'brighter' and cut more. This doesn't have to be a drastic effect; even a little bit of timbral change gives a more dynamic feel.
• Lower the filter cutoff somewhat, and assign it to an envelope generator whose overall amplitude can be velocity-controlled. Using an envelope can change the timbre over time (eg., the decay can be more filtered than the attack). This technique often works very well on longer sounds.
Changing a sample's start point under velocity control is yet another way to add dynamics, although not all samplers offer this feature. Set the initial sample start point several milliseconds into the sample, and add negative velocity modulation so that higher velocities move the sample start point closer to the beginning of the sample. The interplay of modulation amount and sample start time is critical, so expect to take some time to properly tweak this setting. (For more information, see the article by Garth Hjelte in the September 1990 issue of Transoniq Hacker on EPS start point modulation techniques.)
Although the next trick does not affect dynamics per se, it can give the 'feel' of a more dynamic drum sound. Hitting a drum hard stretches the skin tight enough to raise the pitch slightly, which settles back to normal pitch as the drum sound decays. Using a sampler's pitch envelope to give a very slight upward 'spike' to the pitch gives a more dynamic feel, even though this doesn't change the volume. Pitch modulation is particularly effective if you can velocity-control the envelope levels so that only high-velocity hits bring in the pitch change.
Sometimes, what works best is a combination of amplitude, filter, sample start time, and pitch modulation. Don't go overboard with the processing - each element should contribute a fairly subtle difference to the sound, because combination effects multiply the potency of individual effects.
Many samplers let you layer sounds on the same key for 'doubled' effects, and they may also let you switch between the layers according to velocity. For example, a sample of a hard-hit snare could be triggered by velocities over 105, while a sample of a more softly-hit snare could be triggered by velocities 105 or lower. Some samplers include velocity crossfading instead of velocity switching, so that higher velocities fade out the 'soft' sample and fade in the 'hard' sample. This tends to give more seamless transitions between the two sounds. Furthermore, velocity-switched or velocity-crossfaded layers can be processed with the same velocity-related techniques described above.
Sample switching is particularly effective with percussion sounds; try short/long decay cabasas, skin/rim tambourine hits, conga slap/heel, bongo left hand/right hand, etc. Effects such as snare rimshots and hard/soft tympani hits are also well-suited to this technique.
In the studio, many engineers use compression or limiting to decrease a drum's dynamic range. This allows a higher average level to be printed on tape, which gives more 'punch'.
Figure 1 shows a typical bass drum sample; like most percussion sounds, it has a sharp initial attack transient. The problem with recording sounds with sharp transients is that if you set the level low enough so that the peak doesn't distort, the drum's 'tail' will be quite low-level.
The solution requires a sample editor with a gain normalisation option that lets you restrict the dynamics to a user-defined threshold. Set the threshold to about 66% of the existing dynamic range. Figure 2 shows the same bass drum sample, but zoomed in on the attack for clarity; the dotted line shows the threshold.
Now normalise individual peaks to this threshold; the region selected for normalisation should always be bounded by zero-crossings. Figure 3 shows the same region as Figure 2, but with the first two peaks normalised to the threshold. The highlighted section has just been normalised. (It's usually a good idea to normalize an entire half-cycle, but that isn't always necessary if the peak being normalised goes through two zero-crossings.)
Now set the threshold back to maximum, and normalise the entire waveform back to maximum dynamic range. Compare Figure 4 to Figure 1, and you'll see the average signal level is considerably higher than it was originally.
Because this is true digital domain limiting you won't experience the pumping, overshoot, distortion, and other problems associated with analogue limiting. In fact, there's no reason why you can't apply extreme amounts of limiting, although limiting does bring up the noise floor by an amount equal to the number of dB by which the signal is limited. The effects of quantisation noise may become more pronounced as well.
Note that you can also use normalisation to bring up low-level signals, not just to attenuate high-level signals.
Percussive sounds generally have short decays; they grab your attention, and then they're gone. As a result, you may not really need those last several dozen (or even hundred) milliseconds at the tail end of a drum sound, particularly if it contains lots of ambience (which can be synthesised to save even more memory, as described later).
First, move the sample end point progressively closer to the beginning. This will show whether there is any unneeded 'dead air' at the end of the sample, but you'll also be able to determine if the low level signals at the tail of the drum sound are really necessary. Move the end point as far forward as possible without cutting into the sound's essential character, then truncate the sample.
Now apply a fade to the last quarter or third of the sample so that the truncated sound fades out gently instead of just stops (if your sampler can't do fadeouts, a decent visual editor will). If you can lop even just a little space off the ends of your samples, the amount of memory saved can really add up over a full drum kit's worth of sounds.
The lower the sampling rate, the less memory a sample requires. Drum sounds, which tend not to be transposed very far from their base pitch, can often be sample-rate converted without audible sonic degradation. Changing a sound from a 44.1 kHz sampling rate down to 20kHz cuts the memory requirements to less than half. However, this technique does not work well with sounds that have lots of high frequencies - save sample rate conversion for toms, kicks, congas, bongos, etc.
Although most drum sounds do not lend themselves to looping, there is an exception. Toms (unless they include a significant amount of ambience) generally settle down into something resembling a sine wave toward the end of their decay. Since toms often have long decays, you can save quite a bit of memory by introducing a single cycle loop as soon as possible in the sample.
Figure 5 shows a typical tom sound (incidentally, some of the cycles toward the beginning of the sound were normalized to a higher level to give a punchier attack). A little past the halfway point, the sample turns into a stable, loopable waveform. By truncating the sample after the loop, we can reclaim almost half of the original sample's memory.
Of course, looping creates a static, sustained sound that needs some processing. First, set up an amplitude envelope to create a decay/release time so that the drum doesn't drone on forever if you hold down a key. Second, if you listen carefully to an acoustic tom, its pitch will often go slightly flat over time. A tom that changes pitch and then goes into a static-pitch loop sounds pretty bad, so use the pitch envelope to add a slight downward pitch shift at the tail of the tom.
One of the biggest memory-eaters in a set of drum samples is ambience - that tail of reverb that follows a drum sound recorded in an acoustic space. Ambient sounds are much more popular than dry ones, though, so what can you do? Plenty.
The secret is to first create a single, 'universal' ambience sample. Begin by finding a snare sample with a long ambient tail (Figure 6).
To create the ambience sample, locate the point where the primary snare sound has decayed, and fade in to that point (Figure 7). This may require a couple of successive fades to get rid of most of the snare sound. You may also want to normalise this ambience sample for a hotter level.
Next, remove the ambience from all drums (via the truncation/fadeout method mentioned earlier). Figure 8 shows the same snare as in Figure 6, but with the ambience removed and the snare given a fairly abrupt decay.
Assign the ambience sample to a separate layer so that it triggers along with the snare; for the other drums, create copies of the ambience sample and assign them to the keyboard as appropriate.
However, these ambience copies will need to be processed. For a low tom, for example, transpose the copied ambience sample down several semitones and filter out some of the brightness.
Separating ambience can do more than just save memory, since the ambience can be processed into other types of ambience. For example, add an envelope to the ambience sample and shorten the decay to reduce the size of the 'room.' How about gated reverb? Just set the ambience sample's envelope so that it cuts off abruptly. Most samplers also allow you to reverse a sample, giving you access to instant backward reverb.
Another useful trick is to assign the sampler's mod wheel so that it controls filter cutoff on the ambient samples. This lets you 'deaden' or 'brighten' the room.
Separating ambience is a powerful, useful technique that is well worth the effort (incidentally, several manufacturers use this technique in their drum modules). Give it a try.
Another way to save memory is to digitally pitch shift the sample to a higher pitch (this usually requires a reasonably sophisticated visual editing program). The process of pitch shifting shortens the sample time, thus saving memory. Then when assigning the sample to the keyboard, transpose the sample down from the unity (root) pitch by an amount equal to the amount by which it was pitch-shifted up. The result: a sample with the same apparent pitch as the original, but which takes up considerably less space.
Tom samples are often very similar, just tuned differently for high, mid, and low sounds. You can save a lot of memory by using one basic tom sound, copying it to create the other tom sounds, then altering the pitch of the copies.
To add more variety to these sounds so that they don't all seem like copies, use different amounts of filtering and pitch bend. I generally use a high or mid tom for my basic sample, since changing the pitch downward gives a long decay tail which sounds great on low tom sounds (and because it's a transposed copy, it doesn't take up as much memory as a real tom with the same decay length).
There are several ways to make hi-hat sounds more realistic. For example, if your sampler offers a key-up layer (ie., releasing a key can trigger a separate sample, as is possible on the EPS), add a foot-dosed hi-hat sample to the key-up layer, and assign this sample to be triggered by the same key that triggers an open hi-hat sound. Thus, you can 'close' the hi-hat by simply releasing the key. So that you don't trigger the closed hi-hat sample after the open hi-hat sample has decayed, try to find as long an open hi-hat sample as possible.
Also remember that two hi-hat samples should not sound at the same time, since an acoustic hi-hat won't do this. Many samplers allow for a layer with mono response, where playing a new note cuts off any note that is already sustaining. If you assign your hi-hat samples to a mono layer, you don't have to worry about hitting two samples at the same time.
If you don't have enough memory to accommodate both open and closed hi-hat samples, use just the open hi-hat sample. Then, create a copy of this sample, and set the copy's envelope for a very short decay. This will come reasonably close to sounding like a closed hi-hat.
Cymbal sounds gobble up memory, and few samplers include enough memory to allow the luxury of a full-bandwidth cymbal sample with a long decay. Although pitch-shifting and transposing (as mentioned above) can help, there's no substitute for the real thing. So, if you want a real great cymbal sound, just overdub acoustic cymbals over your drum machine tracks. The realism of the cymbals may even fool some people into thinking the other drums are real too.
Due to a gong's extremely long decay time, it's even more troublesome to sample than a cymbal - but there is a workaround.
Transposing a copied cymbal sample way down (an octave or more) gives gong-like sounds, but unfortunately, the extreme amount of transposition will often result in a grainy, distorted attack. However, as the cymbal decays, the sound will appear cleaner because lower level signals are less affected by the effects of transposition.
To obtain a convincing gong, layer the original cymbal sound with the downward-transposed cymbal sound. Set a slow attack time on the transposed cymbal so that you don't hear the initial attack; the main cymbal sound will fill in for it and provide a good-sounding attack. As the main cymbal sound dies, the transposed cymbal will provide the rest of the decay. Because the sample has been transposed way down, it will give a real long decay.
Samplers allow for a number of unusual special effects, such as echo, flams, digital mixing, and the like. The possibilities are nearly limitless; here are a few ideas to get you started.
Aftertouch may not seem all that useful for drum sounds, but there are two main exceptions. Sample a drum or tympani roll, then assign overall level to aftertouch. This lets you bring the roll in and out according to the amount of pressure you exert on the key. Also try assigning filter cutoff to pressure so that heavy pressure increases the brightness somewhat.
Another application for aftertouch involves pitch bending on drum sounds with long decay tails. This allows for 'talking drum' effects, as well as the sounds that occur by applying pressure to the drum skin to change pitch.
Flam effects, where two drum hits occur within a fraction of a second of each other, are often used to add accents to toms and other sounds. The simplest way to do this involves copying the sounds to a be flammed to a 'flam layer', assuming that the layer can be delayed. Set the layer for a delay time on the order of 20 to 50 milliseconds, and you'll have instant flam.
If it's not possible to delay a layer, you have two options. During the sampling process, leave a little bit of 'dead air' at the beginning of the sound to be flammed and layer two of these samples on the same key. For the main sound, move up the sample start point so that the dead air goes away (but don't truncate the sample; just move the start point). For the flammed sound, trim the start point for the desired amount of delay.
If the sound to be flammed has already been sampled, use a visual editor to insert about 50 milliseconds of silence at the beginning of the sound, then follow the same procedure as described above. (I keep a short 'silence' file on my hard disk for such purposes.)
Consider velocity-triggering the layer or sample to which the delayed tom is assigned so that you can trigger the flam by hitting the key harder.
Most samplers allow for mixing individual wavesamples, which lets you create 'monster' composite sounds. For example, once I was searching for just the right kick drum but was frustrated: one had a wonderful-sounding initial attack but was somewhat thin overall, while another had a weak attack but a great, rumbling sustain. A third sample was of a kick drum that was tuned very high.
Mixing these all together created a sound with a great attack, and a tight decay with a strong low end.
How about some echo effects? Loop a percussive sound, with the loop start point at the beginning of the sound, and adjust the loop end for the desired echo period. Process the sound through an amplitude envelope that decays over time, and you'll have 'free' echo without the need for an external signal processor.
The problem with most reverse percussion effects is that they lack the strong attack so necessary for a percussive sound. Try layering a sample with a copy that's been reversed; adjust the mix between the two as desired. This composite sound gives you a sharp attack with a reverse feel because of the way the sound decays.
Incidentally, since layering does require more voices, if you like this effect you can digitally mix the normal and reversed sounds. Since they both last the same amount of time, mixing them will not increase the length of the sample, so you needn't concern yourself with using up more memory. If you want to remove the reverse decay from the mixed sample, you can come pretty close by using an amplitude envelope to form a quick decay so that the sound is gone before the tail kicks in.
You can fatten up most all drum sounds by layering a drum with a detuned copy on the same key. Detuning about a semitone seems to work best; if the detuning is too close, there will be a flanging effect but if the detuning is spread too far apart, the composite sound will be more like two single drums played at the same time instead of one large, fat sound.
Hopefully the above has demonstrated that the combination of sampling and percussive sounds can lead to entirely new avenues of exploration. Have fun, and good luck coming up with the Drum Sound of All Time.
Craig Anderton is an author (11 books), recording artist/producer (nine albums), and also serves as Editor-at-Large for Guitar Player magazine. He has just finished creating a series of sample disks for the Peavey DPM 3 and Ensoniq EPS for Prosonus.
Feature by Craig Anderton
mu:zines is the result of thousands of hours of effort, and will require many thousands more going forward to reach our goals of getting all this content online.
If you value this resource, you can support this project - it really helps!