There is a parallel discussion thread on CA forum for discussions of information in this thread. Alas, I am banned from that forum and folks there don't want to come here. So I thought I address some of the points made there, specifically this one from member mmerrill99:
https://www.computeraudiophile.com/...point-again-looks-for-the-music-in-the-noise/
To summarize, the point is being made that some types of "noise" may be difficult to instrument and see in fourier transforms I have been showing. As an example, a type of distortion called pre-echo which occurs in audio compression is explained and shown in this nice video:
In a nutshell, when there are transients, due to the way lossy compression works in using a block of audio for compression, any "quantization noise" is spread across that block/window. Think of taking some of the energy from the sound of a "tick" and spreading it before and after it. What is after is not very audible (temporal masking) but what is before can be although it requires trained ears to hear it at lower levels. The above video shows this in the case of highly compressed MP3 at 64 kbps with the result of "dirty" sound you get prior to transients.
This is all true. And it is also true that for detecting fidelity of lossy compression, we do not use measurements. Listening tests rule and of course they are conducted blind when it matters.
Merrill goes on to say:
"One thing I mentioned before is that standard FFTs will not show dynamic noise fluctuations of this type so showing FFTs of the analogue output of a DAC is laughably flawed (guess who does this to "prove" these devices have no effect?)"
First forgive me for giving a bit of my resume
but it is necessary here. For a decade I managed the signal processing group at Microsoft among others. We developed a number of lossy codecs such as WMA, WMA Pro, WMA Voice, etc. And I worked closer with this group than any other, routinely working to advance the state of the art and advancing fidelity of lossy audio compression. So the example given above is fully in my wheelhouse.
Now let's discuss the issues with the argument made.
First thing to note that if something changes in time domain, it absolutely shows up in frequency domain. The transform is bidirectional and it cannot be that you add so much pre-echo distortion and frequency domain (FFT) shows no difference. To show that, I went ahead and compressed a standard MPEG reference clip for lossy codecs, castanet, and encoded it into 64 kbps MP3. Pre-echo is very strong there as in examples Merrill shows. Here is the FFT spectrum of the original file (in red) and the 64 kbps MP3 (in yellow) overlaid on it:
At 64 kbps, MP3 has a limit of 12 Khz and see that in the sharp drop at that frequency. That is not material to this discussion. What is material is that at lower frequencies the two graphs do not at all look identical. The curves deviate as they should. The Fourier transform works on a window of time and it easily captured the differences in the original and compressed MP3 with pre-echo.
Interestingly enough, the lossy compression is performed completely in frequency domain! The reason is that our knowledge of psychoacoustics is far stronger in that domain than time. So it reasons that changes that occur in frequency domain would be visible in frequency graphs just the same.
That said, in the case of pre-echo, the distortion is much more visible in time domain. So there is a partial point there but the general assertion is completely wrong that FFT is not revealing of differences.
Even more important here is that we can mathematically show that pre-echo occurs. That is, the objective proof is solid and without doubt at the most fundamental way. There is no such proof that cleaning up USB signal for example changes the analog output of a DAC. That is a leap of faith and there is no math or any kind of objective proof that works in every case as we have in pre-echo.
Here is the most damning point against this argument.
Listeners are completely blind to many lossy compression artifacts. Take even that pre-echo. As obvious as that is, when you ratchet up the bit rate to 320 kbps, 99.999% of listeners fail to hear it and that includes audiophiles! In this case we have ditched 75% of our audio payload (1,400 kbps) and left only a quarter and folks are completely blind to what occurred. Now we are supposed to think that someone is able to sense the cleanliness of USB data pulses in their DAC output? I don't think so.
Now, we can cut through all of this and get to a simple point of agreement. In the case of above pre-echo demonstrations, we can perform the test blind and listeners will absolutely tell the difference between original and compressed. Show me that the same thing happens with UpTone ISO Regen and we are done!
Long time readers of my posts on this topic recall that I accepted a challenge to go to and perform such a test after Alex, the business owner of this company, said he would be able to pass it. But he backed out. My offer still stands.
So until then, let's pay attention to data in front of us. It is not created with any agenda or by ignorance. I understand every argument the designers make and I am here to say that
it is the job of any high performance DAC to not need this device! I like to meet the designer of any high-end DAC and hear him say why he couldn't afford or didn't know about this type of signal cleaning inside his DAC. I am confident there are no such designers willing to do this.