I don't know that this kind of "noise" is what we mean by noise. Correlated noise with music is no longer noise and their analysis should detect that as such. How does the noise sound to you in those youtube videos relative to uncompressed version?
But what is music? Here it is mainly real noise.
Even if we can hear some patterns, I wonder why the noise would be always correlated with it (in most cases it is not).
Often here the noise is spurious, grainy.
Typically noise could be detected (and removed) by determining a magnitude threshold in a frequency-magnitude representation (in a transformed domain),
but this would remove all the low-level texture, noise in these examples of noise music.
Or what special algorithm do they use?
Moreover in some of these examples the noise is the loudest thing... → everything or nearly everything would be erased.
Re. the lossless versions, I do have them, but I think there is prima facie not much difference with youtube, as long as one clearly feels the intended effect of the abundance of noise. Even if youtube cuts a bit of noise, noise is still very strongly present in these examples on youtube.
My point is:
A. In these examples, for at least most of the noise, there no difference between intended and unintended noise.
B. We humans can understand this, because we have use some context (making us know that it is intended noise music).
C. For an algorithm to detect this, it would have to train a machine learning technique on a large corpus of music together with contextual clues to be extracted from the context (e.g. youtube words indicated a noise based musical genre) : something way too heavy to be run afterwards by a compressor like MQA --> it is extremely doubtful, and almost certain, that MQA does not do that. . Moreover given how many exaggerated and false claims they have been using to boost their marketing, if their was some "artificial intelligence"in the process, they would claim it hard and loud... So we have no reason to go into this direction, which leads to the dilemma:
D If MQA would remove the noise, then it would remove most (at least all the fine grained) of the contents of these musics but then that would be really wrong, messing up way too much.
E. If MQA would not remove the noise, then their claim would be false.
My impression:
it is that there is nothing behind their claims of detecting and erasing the noise except a marketing ploy to help people swallow the backwards idea of cutting the 3 LSB of Redbook's 16 bits of audio data, to use the freed space to put instead their DRMs.
In this manner they can not only hide they negative features but also turn them into things falsely looking good, appealing.
So just (misleading) marketing, like their original claims of lossless coding and of not having any DRMs.