Stinius,
I think I probably don't understand what you're asking. Can you clarify?
There is already a "Difference + Reference" feature, where you can re-inject the difference part at any relative gain you want, while listening to the original track in the mix at a fixed level. This is treated as a new track ("Reconstructed Compared Track") that can be played same as if it were one of the others.
stinius said:
I think you already can, 4th row of buttons. Add the diff back into the ref and boost the diff. Unless I misunderstand?
Edit: yes you can, and I can hear the sousa band if I up the diff part 9dB, especially in the low level part at the beginning. With the boost at 6db I don't hear it. But this is at the laptop so possibly on my main system I might be able to hear it earlier. Or maybe not...
Jan Didden
vuki said:
Thank you I found it.
I was only listening to the original signal and the changed signal.
Cheers
Stinius
If I understood correctly; in those 7 sample pairs, one actually doesn't know which the real reference (one without the band) is. And the "reference track" is always track "A"? How come that in any of the 7 sample pairs when I listen to the reference track with +9dB of difference I can hear the band. Shouldn't the audibility treshold (boost) for the sessions when reference track is the one with the band be lower? 
PMA said:
I forgot: in the DiffMaker screen there is a note that the test was with the Z5U loaded with 100k, and that when loaded with 1Meg, the residu was much lower. So it makes a difference with Z5U, and it may well make a difference with the polyprop.
Jan Didden
Noise is in the ear of the beholder! I think there are different definitions of "noise" confusing matters.
DiffMaker is digging out only what is different -- it has no idea what part of a recorded signal we may think of as being the music or the noise (in terms of musical content). From a difference standpoint, noise is something that is always different every time you repeat a test, the part that is never the same even though conditions are identical. Noise that has been recorded into an original music track isn't noise anymore (by that standard, anyway), it is now part of the "signal" since it will always appear the same when you play that recording. But the hiss of an analog preamp playing the signal back right now, though, will always be different from the hiss at any other time. (though it may sound the same to ears, the noise can't be subtracted to make it cancel).
The band is buried in noise on the recording in the sense that the background noise (more the reverb than the noise, but you get the idea) masks the sound of the band so our ears can't pull it out. The band is still there. DiffMaker can pull it out not because it has any better abilities but because it has two recordings, one with the band, and one without.
john curl said:
John, even if your last sentence was correct, which your postings here cast great doubt on, Bob's responses are of a more mature level that your personal attacks. You may not even be aware that by your last sentence you are aiming it directly at Bob. Most of your current postings suggest you have left engineering behind in search of becoming a mystic cult leader. Good luck, but don't expect too many followers here. Joshua will no doubt have a response to make.
bwaslo said:
We can hear well below 'technical level' (visible, visualized) of noise - human's ear selective capability.
bwaslo said:
OK, that's what I had in mind - background noise recorded into original Brahms - part of the signal now. But if you "hide" the band below that noise/part_of_the_signal level of course it becomes inaudible. And of course diffmaker can extract it. Was it a trick to "prove" how much could diffmaker "hear" better than humans?
😕
Hmm, now I'm confused. Are you saying that you can hear the Sousa band in the Reference tracks even without doing a Difference extraction? When you pull the seven different pairs up individually, the Sousa isn't always in the Reference, sometimes its in the Compared.
Reference and Compared are just names. The files could be swapped, with no change in the Difference (except that the Difference from one case would be inverted relative to that from the other). The only difference between Reference and Compared for DiffMaker is that it messes with the gain and the time offset (and maybe the EQ) of the Compared one to try to get the best match, while leaving the Reference as it was.
The "9dB" number is going to be listener and equipment dependent. It depends on how loud you'd normally listen compared to how weak of a signal you can hear on its own (maybe on how messed up ears are from normally listening so loud!). I set the mixed-in level by ear (with headphones) to get it so that I couldn't hear it in the mix but that it would still be embarrassingly obvious, to any golden ear who insisted that his hearing was much more sensitive than the things DiffMaker could expose.
bwaslo said:
No, I do the extraction and then I have the band as difference. And when I boost that difference by 9dB I can hear it in the mix with reference. I think you buried the band in the background noise of the original recording and there is no ear, golden or not, that can hear the band in this kind of mix.
syn08 said:
That means that you measure noise across 20 kHz band, but ear is still able to hear signal burried below this noise level - something like narrow band analysis. Depending to conditions (frequency and level), human ear is able to hear up to 30 dB below noise level.
Similarly, you will record some surrounding noise in the street, e.g. But, human ear in the same place, where was the microphone, is able to detect sound from specified direction, hiden below noise level of the recording.
Well, no, DiffMaker can't actually hear at all. It's just dumb subtraction preceded by some intelligent gain and timing control. Arithmetic, basically.
The point was to head off an argument I was already hearing on some forums (not here, I don't think) that we can all hear stuff too subtle for something like DiffMaker to ever expose. Audiophiles have very high estimates of their ability to hear all kinds of subtle things, this was meant to suggest a little taste of reality.
After working with differencing even a little, it becomes painfully obvious that lack of sensitivity is not it's weak point! Quite the opposite, in fact, the weak point is in its being outrageously sensitive (and hindered) by all the kinds of "normal" differences that no one cares about but which can keep it from canceling out things completely.
And of course part of the purpose was to make a memorable demonstration when the paper was read!
One thing that should be pointed out (and which I don't think has come up yet) is that the Brahms/Sousa recordings were digitally generated and so have none of the noise issues that would hinder a normal test that went through an analog stage.
Well, sure. But the point is that it is THERE and is unheard. And it is quite different from the Brahms choir; and you can hear it in the isolated difference track without turning the gain up at all. And for crying out loud, we're here talking about hearing the changes from using different kinds of capacitors and wire, yet can't hear a Sousa band, complete with drum pounding not very quietly, during a choir recording??
PMA said:
I'm afraid I still don't get it, correct me if I'm wrong. So, if I record through a mike and play the recording in a studio I will hear only the street noise. However, if I am in the street I could hear the signal in the noise.
If I am correct above, what does this gedankenexperiment tell about the ear? Do you have any references about?
bwaslo said:
Well, the band isn't at much lower level than difference artifacts from mp256k sample, and you wouldn't argue (or would you?) there is not so small difference when (audibly) comparing original with its mp3/256 version.
Do you have any explanation for that phenomenon?
- Status
- Not open for further replies.