Creating Personalised AWE32/64 Soundfonts

Using SoundFonts with your AWE32 or AWE64 soundcard makes them totally open‑ended — if you want a new sound for MIDI use you just sample one. Martin Walker guides you through the wonders of personalised SoundFont creation.

If you want to add some personality to your music, creating unique sounds will ensure that you stand out from the crowd. There's no substitute for the hand‑crafted approach, and sampling your own sources can give your track that certain something that makes it different. You don't need a high‑end sampler to be able to do this, as there are various PC soundcards on the market already that allow you to download personal samples, and then play them back, just like any other MIDI sound source. The most popular of these is undoubtedly Creative Labs' AWE32, and their downloadable sample format, known as a SoundFont, has become one of the market standards. The name arises because of the similarity with graphic fonts — each SoundFont contains a selection of sounds that are grouped together into an audio 'alphabet'.

Because so many people have bought the AWE32 card (and its new rich cousin, the AWE64), many companies are now providing ready‑to‑use SoundFonts, giving end users a wide variety of sound sources in addition to the built‑in ROM set already supplied on the soundcard. The beauty of SoundFonts is that the basic samples can also be modified by the AWE's comprehensive synth engine (provided by the EMU8000 soundcard chip). This gives you access to the traditional oscillator‑filter‑amplifier chain so beloved of analogue synth owners. In this case the VCO (voltage‑controlled oscillator) is replaced by the sample, but the VCF (filter) and VCA (amplifier) remain, along with two envelope generators and two triangle‑wave LFOs. This enables you to add much more personality to the basic sound, as well as making it much more expressive to play. However, it's a sad fact of life that when you start talking about creative sampling, or the programming of any sort of synth sounds, many people's eyes glaze over. Most still rely on third‑party products to fulfil their needs, and rarely, if at all, get their programming hands dirty. But, despite what you might think, creating your own SoundFonts does not require a white coat, a library of manuals and copious supplies of black coffee. Once you know the principles, you could have a basic SoundFont completed using existing WAV files in an hour or less, and this should spur you on to discover more.

Getting Your Feet Wet

There are various ways of using SoundFonts. Some people prefer to load a batch of instruments into their soundcard RAM, each occupying a modest amount of memory. In this way you can create a whole ensemble for a particular track (sample CDs of ethnic instruments lend themselves very well to this sort of technique). Or you can blow your entire RAM allocation on one memory‑hogging instrument that really lifts a track, which otherwise uses only the basic ROM sounds. Remember that many professional musicians working with orchestral sounds and big budgets find that they can achieve more realistic and natural sounds by layering one real recorded instrument against a batch of synthesized ones. At the lower level of PC soundcards, this technique can still work. When the bulk of the available sounds are made up from very short loops, having a larger sampled lead instrument or evolving pad will give a track much more lasting interest.

However, you don't have to start amassing expensive studio mics to capture sounds from the real world. Audio sample CDs provide a huge pool of sounds, but these don't necessarily need to be plonked into place just as they are. Once a sound is in WAV format in your PC, you can subject it to all sorts of treatments using one of the many editors available. By adding straightforward multi‑effects to individual samples directly, you can achieve mixes otherwise only achievable with multitrack. More bizarre techniques, such as reversing or using extreme distortion or quantisation effects, can also create new and original sounds, and these can then be saved as new WAV sounds before being incorporated into the SoundFont.

Many bands work almost entirely with samplers nowadays, treating their sounds in so many ways that few would recognise their original origins. You can often find musical inspiration from the resulting sounds as they spark off some fresh train of thought. A completely different approach is to create the original sound from scratch, and there's a wide variety of software manipulators and generators available (very well covered in the recent two‑part SOS feature 'Making Waves', in the April and May issues), ranging from shareware to full‑blown commercial applications supplied on CD‑ROM. These are another ideal way of providing unique sounds and textures.

Back To Basics

Supplied with the AWE32 and 64 soundcards is the Vienna SoundFont Studio, which Creative Labs term a 'Professional Preset Editor'. This has evolved over the years, and the latest version 2.1 (which is shown in most of these screenshots) is supplied with AWE64. Details on how to upgrade are given in the 'Oh Vienna!' box. To create your own SoundFonts, the first step is to select 'Import User Samples', to load in some of your chosen sounds in WAV format. Figure 3 (overleaf) shows a stereo percussion loop being imported from the excellent Spices Of India sample CD (part of Time & Space's Creative Essentials series). Once imported, this stereo source will appear as two separate samples. In Figure 4 (overleaf) you can see that the User Sample Pool now has two entries — the stereo percussion sample has automatically been split internally into left‑ and right‑channel samples after being imported, and its Pan values have been set to hard left and right respectively.

It really is a case of 'click, drag and experiment'.

Once you have samples in your Pool to choose from, create a New Instrument by right‑clicking on the Instrument Pool (Figure 4). The Instrument uses the synth engine mentioned earlier, passing the sample through a filter and amplifier, and allowing various modifications and treatments to be carried out en route. Once you've chosen a name for the new instrument, you can decide to go to the New Instrument Zone, where you select which samples will be included. In this case, it's a stereo sample, so both the left and right channels need to be chosen (you hold down the Shift key while clicking on them to select both simultaneously). Each zone consists of a sample with an associated note range. To adjust the note range for a particular zone, first select it, by clicking either your left mouse button on the appropriate horizontal line beneath the keyboard, or directly on the sample name in the Instrument Pool. Once it's selected, you can change the note range for the sound by left‑clicking and dragging either end of the current zone symbol, or by right‑clicking and selecting 'MIDI set range', and playing a low and high MIDI note on your external MIDI keyboard (see Figure 5, page 124).

If you want your samples to loop, you'll have to select this for each item in the Instrument Pool, using the right‑click and 'Loop...' option, which brings up a waveform plot and a looping adjustment and audition section. This allows you to set loop points, but it is pretty basic. While you're in the instrument section, you can have a lot of fun altering all the synth settings — it really is a case of 'click, drag and experiment'. With the new instrument basically defined, you will finally need to create a New Preset, in order to give it a MIDI bank and Program number so that it can be easily accessed by a sequencer from the outside world. Once this has been done, you can save your completed SoundFont bank using the 'Save As' option from the File menu. Now you're ready to use the new SoundFont in MIDI sequences.

To download your new SoundFont into the soundcard's RAM, load up the AWE control utility (see Figure 6, page 124) and then select the User page. Here you can point to your new font (using Browse is the quickest way to find the newly saved font), and, once the path has been set up, click on Apply to actually download the new sounds into the User bank. They can now be accessed as easily as any other MIDI sound, and used as part of your music compositions.

Multisampling

Many people are put off creating their own SoundFonts by their apparent complexity, but just because you're presented with a designer kitchen doesn't mean that you can't make beans on toast. The easiest way to start is by mapping a single sample across the keyboard range, as above — although many real instruments can only be played over a comparatively small span of notes, so there's no point in trying to create perfect results over seven octaves! Even if the instrument has a massive range, it is unlikely that most music tracks will need to use it all. Having said that, one of the benefits in only using a single sample is that you can often find unexpected but interesting results if you use it way out of the normal pitch range, which may spark off more musical ideas.

Multisampling involves using a set of samples, each optimised for a particular pitch range. By splitting up the sound in this way you get a more realistic result, as most real instruments change in timbre over their range. The easiest way to try the multisampled approach is to grab some of the many sounds provided on CD audio sample disks (and in WAV file collections). These are often provided as a set of three samples, at C3, C4, and C5 MIDI note values, ensuring a reasonable note range without too many bizarre changes in timbre. Simply allocate each one a range of one octave, and then stand back and admire your handiwork. For many instruments, the transitions may still sound quite odd, and the next step up is a half‑octave multisample set, which has, for instance, C3, F3, C4, F4 and so on. Many ROM sample sets go one further than this, if there's space, by providing 1/3 octave multisamples, with changes in sample every three or four notes. You can examine the contents of your ROM set in the Vienna editor, as individual ROM samples can be imported, and multisamples include their designated pitch in the name. The number of multisamples used varies from one (for sounds such as ocarina) to eight (for the piano, as you might expect).

Cramming It In

The AWE32 comes with 512k of sample RAM already on board, but if you are using GS mode this is already used to store the sound effects, and the MT32 mode also uses a small amount of this memory. GM mode leaves the entire 512k free for your own use. Although some people have AWE32 cards with only the fixed 512k, this can be expanded by up to 32Mb of SIMM memory (although 4Mb is always mapped by the card as ROM space, so only 28Mb of this will be available for your own samples). A fully expanded 32Mb soundcard contains as much memory as you'll find on many rackmounted industry standard samplers, so you're unlikely to have to use any tricks to fit in a good range of sounds. However, it's surprising how many people neglect to remove unwanted sections from their samples. If you're looping a sound, and there is still data after the loop end point, this can normally be cropped. If you have loads of memory, it's easy to be lazy about this, but sooner or later you will fill up the memory and be looking to gain more space. Removing the excess now can save you a boring job later, and instantly regain you some sample space.

If you have only a few megabytes or less, or are interested in creating much longer, evolving sounds that wouldn't normally fit in existing memory, various tips and techniques can be called into play. A favourite way of fattening up short sounds is to layer them. Vienna allows many different samples to be triggered by the same note, and this can be used with dissimilar sounds to create interesting composites. One of the beauties of layering is that if each sound has a different loop length, they will not remain synchronised, but slowly move apart, resulting in a loop length that's effectively a lot longer. Incidentally, one of the limitations of the previous Vienna, version 1, was that stereo samples were not supported — you could load each channel of the stereo sample separately, but unless you were very careful with loop lengths, the two channels would drift apart. Although versions 2.0 and 2.1 allow stereo samples to be directly imported (see the Vienna box) , looping must be set up for each channel separately, so you will still have to be careful. Often the solution is to apply a little lateral thinking. If you want a stereo drum or percussion track, it's easier to use a mono drum loop, and then add extra percussion, panned left and right to give a good rhythmic spread.

Although running two completely different sounds simultaneously is one approach, another classic technique is to use the same sample twice, but detune each version, one slightly up in pitch, and the other slightly down. This will create natural 'chorusing'; if you want a stereo spread, simply panning the two sounds hard left and right will give you a good wide image that still remains reasonably mono compatible. Bear in mind, though, that if the two sounds are not sufficiently separated in pitch, they well sound 'flangy' when run in mono.

You don't have to restrict yourself to single notes when creating new samples. Complete ambient backdrops also work well, even using sounds transposed down by several octaves (and therefore lasting much longer), and this is another way to make the most of limited sample memory. If you think that 512k is too little memory to work with, you've only got to load up some of the demos provided with the AWE series cards to change your mind.

Looping The Loop

The time‑honoured way of creating a much longer sound is to loop the sample, but unless the loop itself is quite long, the sound will lack interest. Short loops are a common flaw of the onboard ROM sounds fitted on many soundcards, but often these do sound rather better when heard in context, with a variety of other instruments. Trying to achieve smooth and glitch‑free loops is one of the things that dissuade many people from producing their own samples. One way to avoid it altogether is by using shorter single‑shot sounds and bringing in other MIDI sources for those longer evolving pad sounds. If you want to blow all your memory on one big sample, on the other hand, you can probably get by without having to loop it. However, looping can save a lot of memory, and it's really not that difficult once you get started.

Most modern audio editors will allow you to look for zero crossings (where the waveform has a value of nought). Making sure that any sound starts and ends with a zero value will help you avoid stray clicks, but for the same reason, this is also the best way to choose loop points. It can take a long time to find the perfect loop point, and patience is definitely required, but you'll soon get a feel for which sounds are more suitable for successful looping. Sadly, very little software does more than jump between adjacent zero crossings, even on commercial rackmounting samplers, and this is why it can still take a long time to find a suitable loop point — although each end of the loop has the same sampled value, they may have totally different timbres, giving an audible jump in tonal quality every time the loop point is passed. On many sounds this means that the only successful loop will be found right at the end of the sample, with a fairly short length. With percussive sounds, most of the evolving part has finished by this stage, but you still get the effect of a much longer sound by continuing to loop round the very last bit. On some occasions, even a single cycle of the waveform may be sufficient to provide a stable (albeit boring) loop. Vienna specifies a minimum loop length of 32 bytes if you want to try this.

To achieve longer loops that are still smooth, a popular technique is crossfade looping. Although this is possible to do by hand, it's a hell of a fiddle, and is far better achieved by a dedicated software function. One of the best editors with this function is Sound Forge, but this is, unfortunately, quite expensive. Crossfading only works effectively if you already have a reasonably smooth loop. Once you have this, a portion of the waveform to either side of the loop is copied, reversed, and added back into the same place. The results, in the before and after portion around the loop point, fade smoothly into each other, rather than just butting up as they did previously. Crossfade looping can solve a lot of tricky loop problems, and many people swear by it, but it does have the disadvantage of permanently altering your sample data, and the crossfade will still give a cyclic change in tone each time it loops. For this reason, it's best to use longer loop lengths, sufficient to make this cyclic variation sound more like vibrato, rather than a rapid 'wobble'. It will also mean that if you try to transpose the sample significantly upward, the speed of the loop will increase, and again reach a point where it becomes unuseable.

In the case of our looping rhythm section example, although we could trigger the sound once and leave it looping continuously during a long section of music, given the amount of other things a PC is normally asked to do it's safer to set up a single shot of the loop in the sequencer, by adjusting the note length to play the whole loop length exactly once and then copying this as many times as you need it during the track. This way, the sample gets re‑triggered in sync with the rest of the music every time, and no cumulative timing drift can occur.

Don't be put off by some of this apparent complexity — most people start off by producing SoundFonts containing just a single instrument, and build up from there. Something about little acorns and mighty oaks springs to mind. Happy sampling!