wlangfor@uoguelph.ca wrote:lalalandsynth wrote:Quick look , it seems that the gain manipulation is done with greens and a 100 tick , there is a response time problem with that , and I think it would vary . Also , faster then realtime rendering would not sound the same and be random basically.
Not really, because there is a tick 100 and averages used, the intermittent variance just serves for the averaging to smooth the signal. A way of using Math to be a peak limiter in a minute sense.
Lalalandsynth is correct - the method is fundamentally flawed mathematically. Without including the contribution of the samples which you skip, an accurate reading of the audio signal's power isn't possible.
Essentially it's resampling the audio at a lower sample rate, so the Nyquist Rule applies - frequencies higher than half the sampling rate cannot be accurately represented (in this case: f > ~50Hz). Any frequencies higher than this will become aliasing noise, and with such a low sampling rate, much of that is going to be at such low frequencies that it will still affect the output averaging (which is effectively a low-pass filter.) During a non real-time render, the down-sampling ratio would effectively be higher, compounding the problem.
Another way to look at it is this. If we assume for a moment that the ticker really does run at exactly 100Hz, what would happen to a waveform also of 100Hz (or a harmonic thereof)? It would depend on the phase relationship between the two - we might capture a series of peaks, a series of zero crossings, or somewhere in between - completely at random depending when playback started (and ticker jitter). As whole series of samples would be biased by the phase relationship, post-measurement averaging won't compensate for this.
Such errors might average out for a sufficiently broadband signal, but (un)fortunately, music relies on discrete relationships between frequencies; so in practice, this is highly unlikely. Hence the averaging must always be done over every single audio sample BEFORE sampling the resulting average, whatever the output sample rate - averaging AFTER sampling simply won't give the same mathematical results.
This can be seen clearly by feeding an oscillator into the plugin and comparing input to output on a scope at different frequencies - you can clearly see the output amplitude wavering, even though the input has a fixed amplitude. It also won't help that you're modulating the output level with a green signal - every tick is effectively producing a discontinuity in the waveform, so you're adding noise, too.
wlangfor@uoguelph.ca wrote:And now even humbly I can say that this thing is one of the best VST plugs ever made
[NB: Personal opinion, not a moderation message.]There is nothing "humble" about such a claim. This is a forum for the kind of people who
will run your plugins through a scope and frequency analyser, stick every kind of signal they can imagine through them, and have a good old poke around inside. Hence, posting your plugins for objective criticism is a good route to learning our vocation better. But your constant hyperbolic marketing up-talk is becoming very tiresome and is not relevant here - please save it for your personal shop-front website.