Thankfully, as you say, even if not always understood or adhered to, at least the need for dither has been widely disseminated and is well known these days, and well supported by the available tools.Psychoacoustic treatment is not limited by using masking. Dithering is a signal processing that always results in degradation of a signal waveform (adding noise) but the same time it improves auditory perception of that signal. Such “magic” processing can be based only on some knowledge of perception - a product of psychoacoustic research. So, dithering is a perfect example of psychoacoustic treatment of audio signal. I think it is obvious.
Anyone can take your piano samples and sort them according to closeness to the original. Having a number of such “sortings” performed by various people, will they all be identical? Especially if some of that people don't like piano and love overdrived electric guitar ... or some of that people never heard of piano and for them it is just an unknown sound, which they hear for the first time. That is why some relation between measurements and perception can be found only for simple signals, in such cases perception is just an acquisition of sound signal without comprehension. In case of complex music signal the comprehension matters and the relation becomes extremely complicated, it can not be reliably found in practice (even by means of neural network). The solution is different.
Thanks JJ.
The question one might ask is, what about the third order error, and why can it be (perceptually speaking) disregarded?
INTRUDER ALERT INTRUDER ALERT!!!!
How about listening to a recording of the real hardware (not Mega Drive though) and read the comments from one of the authors of these emulators...
Oh dear, metallic FM synthesis "clang clang" patch alert!
Trying to keep "on topic," all the "oddities" of the old FM synthesis chips need to be considered if you want an "accurate" emulation. IIRC, I think someone uncapped the OPL3 (horrible low cost 2-op IC!) to document what was really going on inside?
We get onto which "artifacts" are actually a desired part of the instrument's sound. The really severe truncation that happened in the MegaDrive hardware, surely not? I suspect the emulators, if operating in float internally, aren't dithering their outputs, though?
If you want the cleanest possible FM synthesis, then it needs to be running at the highest possible sample rate to avoid aliasing, as modulating one operator with another creates an infinite series of partials/harmonics. Thus, it's not like generating a sawtooth or square wave using band-limited techniques.
In the world of DSP, sufficient bit-depth (in processing) and dither alone are not the only things needed to avoid complete garbage being spewed out!
I think that comparison of error signals with one another results in more useful results than comparison of error signal to original. Art.signatures actually represents (in some sense) error signals computed with vast amount of real music material. Their comparison (in the form of distances) helps to separate different types of signal degradation into clusters according to their similarity. Within each cluster correlation of perceived audio quality to df levels is high enough [https://www.audiosciencereview.com/...od-for-measuring-distortion.10282/post-285578].
Yes, df metric accounts ALL. Its measurements are simple as volt measurements. It can not be used as is for predicting/assessment of perceived audio quality. It can be used for the purpose only in combination with art.signature analysis, which deals with the knowledge of perception, though in a very simple way.
Yes, I can do sometimes as English is not my native tongue. That is why I use sometimes two words (slashed) instead of one.
I used the term "digital silence" in a way as it is used in Audio Precision waveform generator - dithered silence, not equal to zero.
