The question on the mailing list was: why does my synth always sound different after loading a patch or after the G2 recalculates. Rob Hordijk gave the answer:
The basic idea of 'virtual analog' is to not always have exactly repeating sounds, just like samples that will simply always sound exactly the same.
Instead, the idea is to get roughly the same sound, but with the slight changes from note to note that characterize acoustic instruments and analog synths. To do this in a convincing way is quite an art, an art that each of us has to learn over time.
This issue you mention has in general to do with how digital oscillators have a defined pitch but an undefined phase. Small shifts in phase between the waveforms of two or more oscillators can produce drastic changes in timbre. This is in fact one of the hallmarks of analog synths, they do never sound the same as well. Samples however will always sound the same, too much the same in general.
What one needs to keep in mind is the difference between digital and analog oscillators. Analog oscillators always drift constantly, so the phase relation between two or more oscillators is constantly changing, creating a chorus effect. Digital oscillators do not drift by themselves, so when using more than one oscillator they should always be detuned slightly by two to ten cents. This will recreate that lush and lively drifting effect of analog oscillators. But if the digital oscillators are not slightly detuned, every voice will appear to have a different timbre and after every recalculation of the patch the timbres for each voice will change. This is simply because the phase relation stays fixed during the sounding of a note if the digital oscillators are not purposely detuned.
The only way to prevent oscillators to have undefined phase shifts is to use hardsync, meaning that the output of the lowest tuned oscillator is connected to the sync input of oscillators tuned to a higher pitch. But this will also give the typical sonic effect of hardsync, and might not be what one wants.
So, the rule is to always slightly detune the second and third oscillators in a voice, or else use hardsync.
A good trick to mimic the ever and randomly driftting analog oscillators is to feed them all a slowly varying smooth random LFO signal to their pitch modulation inputs and open the modulation knob just one tick. Of course each oscillator must have its own randomly tuned random LFO waveform. This will give different sounds on each keypress and also after each recalculation of the patch. But it will be a dynamic and natural effect, immediately giving life to the sound.
An example to illustrate the above:
In addition to my previous reply a simple two oscillator patch that shows this detuning thingy. Both oscillators have a clocked Rnd module connected to their pitch inputs. These Rnd modules are clocked by the oscillators themselves, so each period of the waveform of an osc is slightly different.
If you mute the two Rnd modules the static character of the oscs can be clearly heard. But if both Rnd modules are On the sounds get very lively.
Mostly lower notes need a little deeper modulation as higher notes. This is done by applying the Keyboard morph on the pitch inputs of the oscillators.
/Rob
