|
cpu, bus, and fp sizes aren't there yet.
24bit/96khz is plenty for recording and fx application.
My experience with 32bit / 64bit floating point aka decimal numbers--assuming our circuit emulations are floats and not fixed integer aka whole number math--is that when you add a small amount to a large amount you get a massive error. In the grand scheme of things these errors don't really matter. But in the narrow scheme of things, these errors (I think anyway) are what cause software to sound weak in comparison to analog--that 5%.
Analog is also continuous and perfect. The performance of analog circuitry may be massively imperfect--a circuit on paper may produce different results in reality as time goes on and components degrade or interfere with each other. Throw in noise and tuning issues (one of my favorite you tube videos is a xoxbox going out of tune because of a candle placed nearby) and I'd guess there are a lot of parameters that go into circuit emulation.
Even @ 192khz I think software emulations are missing on huge swaths of data, because sound @ point Z is dependent on the cumulative results of points A - X. (Making this up), if current volt = sin ( cos (1/tan (prev voltage)), the difference between 32 bit floats and 1024 bit floats is dramatic (1024 may be overkilll), and @ 2 million samples per second (vs a paltry 192 thousand) even more accuracy.
When we hear a softsynth and go hmm that sounds weak, what we are really hearing is not poor programming (per se), but instead errors within the emulated circuit, due to an inability of the computer to contain the sound we (I) want emulated. A simple knob w/a simple knob tweak pot leverages many organic properties of our universe as we know it. The poor little computer only knows on and off--any property has to be programmed, and I don't think 64bits @ 192khz is accurate enough to do say the squelch of a tb303 or the rumbly bass of an aging sh 101.
There should be a calculated point, perhaps specific to each circuit, where the ear simply will fail to hear the result of any imperfect circuit emulation, and 1024 / 2Mhz may be overkill. If c-sound had variable math and sample rate that would be an interesting test, but Alas I fear it does not. Right now, at high resonance and overdrive, I have yet to hear a vst synth that did not sound digital. Sure it sounds good when you are itb, and when I listen to xm radio I really couldn't say which is itb and which isn't; all the same, when I wander over to any analog synth and hit a note, it is still night and day (for me).
|