multisempl [Arhiva] - YU-MIDI Muzički Portal

basalski

30.03.2008, 03:17

Ja mislim da je to drugi rec za; WAV FILE

; ovde ti sve na engleski pise sta je multisample;

Multisampling
To overcome all these transposition-related problems, the concept of 'multisampling' was devised. Ironically, this involved returning to a situation more like that on the pre-digital Mellotron, where each note had its own associated tape recording. When multisampling an instrument, you make recordings of the instrument playing at several different pitches across its range, and then map the right ones to the appropriate keys of your keyboard. Fortunately, with decent transposition/interpolation algorithms, you don't usually have to go as far as taking a sample for every note, although you can if your sampler's memory allows this. It depends on the instrument you're sampling, and how realistic you want the sampled instrument to sound. With some instruments, two samples every octave — on 'C' and 'G', say — might suffice, although you might notice envelopes and other characteristics of the sound speeding up and slowing down as you move further from the original pitches of the samples. Of course, in these days of plentiful memory, it's tempting to have a separate sample for every note, although as with my comments in the box on looping a couple of pages back, I think such 'modern' practices make for very memory-intensive sample libraries, which can take a long time to load, and which can put a lot of strain on the host CPU in the case of software samplers.

One demonstration of why velocity switching is necessary. Shown here in a stereo editor are three separate performances of the same combination of notes (three G#s in different octaves), each louder than the last, from Alfred Brendel's classic 1972 piano performance of Movement 3 of Beethoven's 'Moonlight Sonata'. As you can see from the shape of the waveforms, even the same notes played more loudly or quietly on a grand piano produce very different waves, with the louder notes creating more high frequencies. You certainly can't achieve a realistic sampled piano sound by sampling the notes at one velocity, and making these samples play back more loudly with increasing velocity.
When multisampling, I find a good compromise is to have a sample every minor third in every octave the instrument covers. This way, a sample is never transposed more than one semitone up or down, so you're unikely to hear any serious 'munchkinisation' or transposition distortion, and for the vast majority of instruments, this will be more than enough to give an accurate representation. Admittedly, if it's an instrument that will be featured in isolation, such as solo piano, this might not be sufficiently realisitic, but in a mix with other instruments, you'll find that this compromise works well enough even with potentially troublesome sound sources like the human voice, acoustic piano, saxophone and so on. From a practical and technical point of view, an instrument sampled this way is also going to be a third of the size of the same instrument sampled on every note. Even if you're not doing your own sampling, it can be a worthwhile exercise going through some of the enormous sample libraries currently available and trimming them down, using samples only every minor third to create your own 'lite' versions. These can be useful for situations where absolute realism isn't required — for use in a busy mix or for when playing live, for example.

Whilst recording multisamples seems to make a great deal of sense, it brings with it many considerations, not least of which is maintaining an even playing technique for each of the notes you sample. Obviously, if some notes in your multisample were played more loudly when they were sampled than others, then the sampled instrument will sound very strange when played up and down the keyboard, with different notes potentially sounding louder than others. Actually, overall level is less of a problem, as samples can all be normalised or otherwise level-balanced in the sampler once they've been taken, but the source instrument's tone will invariably be brighter or darker when played more loudly or softly, so it is important to play the instrument such that the recordings making up the multisample sound consistent.