EXS 24 Logic Sampler Users Group

"Hector" <hector@...> wrote:
>>>Hector wrote:
>>> Create a sample by bouncing a note playing middle C..  You only need 1
>>>cycle
>>> of the waveform.
Sasha wrote:
>>Well, you better bounce like C0 through C6, a) to reduce CPU usage (which 
>>is noticeable more or less drastically), especially on chords and b) to 
>>avoid aliasing (which is indeed happening once the EXS has to pitch zones a 
>>lot).
Hector wrote:
>Maybe I wrong, it's a long time since learnt about this,  but I always 
>thought that aliasing happens when the harmonics of a wave pass up and 
>beyond the 20KHz frequency limit set by a 44.1 sample rate.   As a pure sine 
>wave has no harmonics I do not see how aliasing could occur in this case.    

That is the situation which applies when digitally recording analog
waveforms which contain frequencies above the Nyquist Limit (22.05 kHz in
the case of 44.1kHz sampling) -- anything above the limit will 'foldover"
so a 21kHz wave will foldover as a 950Hz wave. This spurious tone is an
alias and the sum of all these tones in a recorded signal is "aliasing noise" .

Sampling is a different matter. To take a set of digital samples
corresponding to one note -- C3 say -- and  spread it all over the keyboard
there are a couple of obvious approaches. First, you could use the same
sample set and just speed up the rate at which you read it -- read it at 4
times the clock speed to play it two octaves up or 1/4 the clockspeed to
transpose it down two octaves -- this is a bit like speeding up and slowing
down a tape recorder. 

However  -- if you want to preserve the same sample rate and audio fidelity
then you need to construct a new set of samples for every trasposed pitch.
Eg -- to go two octaves down you need to interpolate 3 samples (computed
guesses) between every sample in the original note.  To go two octaves up 3
quarters of the original samples need to be removed -- the process is
called decimation.

The quality of the algorithms for this interpolation in critical and all
samplers are not equal. It is this process of interpolation and decimation
that produces waveforms with frequency content not consistent with the
frequency spectrum of the original sample. Subsequent operations on these
waveforms by filters, pitchbend, glide and  FX also have potential to add
aliasing noise to tones.

>Maybe you think the sine in question is not very pure?
Doesn't matter how pure it is if it is going to be put through a process
(intepolation/decimation) which pollutes it.

As an example RGC Audio (Pentagon 1) uses wavetables as sources for
oscillators -- the sine wavetable is one cycle of a sine wave for every
midi note 0 - 127 -- starting with perfect sample sets with no aliasing is
the approach taken here to avoid aliasing in the output. Recording a bit of
C3 and playing it all over the keyboard is going to sound worse than
sampling every note or more practically every octave. 

Regards,
Murray