You can produce more even-sounding multisample patches by using the Akai S5000's Keygroup Crossfade function.
I have recently bought an Akai S5000 sampler and I am attempting to create samples of some synths I have. Getting the samples in is fine, and so is sample editing. I understand the concept of multisampling and I am taking a sample every octave, but when I map them out across the keyboard I can hear very noticeable tonal changes between the keygroups. How can I overcome this?
Andy Blakey
Sound library developer and SOS contributor Steve Howell replies: You're on the right track with multisampling, but one sample per octave is probably not enough in this case. Even with multisampling, samples are still subject to 'munchkinisation' — that is, a change in tonal quality as a result of speeding up and slowing down the original sample when you play different notes/pitches.
You don't say which note you are sampling for each octave or your keygroup low/high notes, but let's say you're sampling the F of every octave with a keygroup range of Cn-Bn. What's happening is that at, say, B2, your F2 sample is transposed up six semitones and so is sounding a bit bright and thin and, well, 'speeded up', but at C3 right next to it, your F3 sample is transposed down five semitones and is sounding a bit dull and, well, slowed down! Juxtaposed next to each other , as you play across keygroups, the tonal jump may well sound abrupt. So what's the solution?
Of course, the most obvious one is more samples per octave, so that none of them is transposed too much to exhibit these 'munchkinisation' effects — maybe the C and G of every octave with keygroup mapping from A#-E# and E-B respectively so that at any time, the samples are only transposed by two or three semitones up or down. Of course, even more samples would be better — sampling every minor third (C, E#, F# and A of every octave) would mean that samples are only ever transposed by plus or minus one semitone. However, this brings its own problems, not least of which is that there are twice as many (or more) samples to edit, trim, normalise, loop and so on. So what's to be done for a quick fix?
Firstly, I would head for the S5000's Keygroup Crossfade function. By overlapping the keygroup ranges and enabling crossfading, at the crossover point of the keygroups you would hear both samples, with each one merging into the other — this can actually work quite effectively for smoothing out the transition between adjacent keygroups. The only problem you can run into is chorusing and/or phasing artefacts at the crossover point if the adjacent samples aren't quite in tune. That said, a judicious tweak of each keygroup's tuning can temper this.
Another technique you can try is to experiment with the upper and lower extremes of the keygroups. One's first assumption is to have the sample somewhere in the middle of a keygroup (for example, samples on F, keygroup ranges of Cn-Bn). However, depending on the nature of the sound, this might not always be appropriate. For example, you might have a sound that transposes up well but not down. In this case, try a keygroup range of, say, E#n-Dn+1 so that the sample isn't transposed down as far but extends further up. The converse could also be true.
Of course, the only real solution is to record as many samples as you can across the instrument's range (ultimately, one for every note!) and dispense with the hassles of looping by simply recording much longer samples than you will ever need to sustain. When I started making sound libraries back in the '80s, entire pianos and orchestral string sections had to fit into 2MB of memory and be stored on 1.4MB floppies. These days, these limitations simply don't apply with the latest raft of hardware and software samplers that feature either large amounts of RAM and/or disk streaming.